University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
ADE | GroundImprovement Depthofinfluenceofrollingdynamic
compaction
Scott,JaksaandMitchell
Number of blows per 50 mm penetration Table5. Valuesofkfordifferenttowingspeedsbasedon
changeinpotentialandkineticenergies
0 1 2 3 4 5 6 7 8
0 v:km/h mgh:kJ ΔKE:kJ mgh+ΔKE:kJ k
0·1 9 11·8 10·0 21·8 1·8
10·5 11·8 13·6 25·4 2·2
12 11·8 17·8 29·6 2·5
0·2
v,speedoftowingunit;k,ratiooftheenergyimpartedtothegrounddivided
0·3 bygravitationalpotentialenergy
0·4
and avariable k, which depends on the towing speed, as per
0·5
Table5.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0·6
k2n2
5: EDI¼ ðmghÞ
g
0·7
0·8
Third, for determining the maximum layer thickness that can
0·9 Zero passes Eight passes 16 passes be compacted in thick lifts, the concept of the depth of major
Figure9. DCPtestresultsforzero,eightand16passes improvement (DMI) is appropriate. This applies to situations
where a target criterion that is comparable to what can be
achievedbyconventionalcompactionequipmentinthinliftsis
Table4. Predictedchangesinpotentialandkineticenergiesfora required.ConsistentwiththedescriptionadoptedbySlocombe
towingspeedof10·5km/h (2004) to determine the zone of major improvement from the
v:km/h v:m/s v:m/s v:m/s ΔPE :kJ ΔKE:kJ EDI, a reduction factor, r, is used. DMI is equal to r (a con-
i f g
stant that varies between 0·5 and 0·67) multiplied by the EDI,
10·5 2·92 3·21 2·63 11·8 13·6 asdefinedinEquation6.
v, speed of towing unit; vf, module velocity after impacting the ground; 6: DMI¼rðEDIÞ
v,modulevelocitypriortoimpactingtheground
i
Values for EDI and DMI are summarised in Table 6 for
differentvaluesofk,ascalculatedinTable5,andn,consistent
Working within the limitations of RDC ensures that quantifi-
with the range of values proposed by Mayne et al. (1984).
able improvement occurs and the properties of the ground are
Lowervaluesofnareapplicableforclaysoils;highervaluesof
improved such that a specified target criterion is met. The
n are valid for granular soils; mixed soils require intermediate
concept of an effective depth of improvement (EDI) is most
values of n to be adopted. The calculated values in Table 6
relevant for applications involving improving ground in situ
are in broad agreement with the case studies summarised in
(as per the case studies referenced in Table 1). The EDI can
Tables1and2.
be considered as the equivalent of the term described by
Slocombe(2004)fordynamiccompaction,beingthemaximum
For the field trial described in this paper, RDC was measured
depth to which significant improvement occurs. As shown in
to have an influence at a depth of 3·85 m; however, the
Equation 4, the new parameter EDI is calculated as the
majority of improvement occurred within the top 2·0 m from
product of Equation 2 (based on module mass, m, lift height,
the surface, consistent with the definition of the EDI. While
h, and empirical factor n from dynamic compaction theory)
RDC improved the soil beneath this so-called effective depth,
and a new term k, defined as the ratio of the energy imparted
for a uniform soil profile, the magnitude of improvement
to the ground divided by the gravitational potential energy, as
beyond this depth was less significant. A maximum dry
listedinTable5.
density ratio of 95% with respect to modified compaction was
pffiffiffiffiffiffiffi obtained for a layer thickness of 1·2m (DMI). The values for
4: EDI¼kðn mhÞ
EDI and DMI obtained are consistent with Table 6 for an n
value of 0·8, reasonable for granular soils, and a k value of
Alternatively, Equation 4 can be re-written as shown in 2·2, consistent for the 10·5 km/h towing speed adopted in the
Equation 5. In this form, the EDI is written in terms of the trial. Table 6 suggests that the depths to which RDC can
material characteristics, n, gravitational potential energy, mgh improve and compact granular soils is influenced more by
8
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license
m
:htpeD |
ADE | GroundImprovement Depthofinfluenceofrollingdynamic
compaction
Scott,JaksaandMitchell
Table6. PredictedeffectiveandmaximumdepthsofimprovementforRDC
v:km/h n m:t h:m D:m k EDI:m r DMI:m
9 0·3 8 0·15 0·33 1·8 0·59 0·5–0·67 0·30–0·40
9 0·5 8 0·15 0·55 1·8 0·99 0·5–0·67 0·49–0·66
9 0·8 8 0·15 0·88 1·8 1·58 0·5–0·67 0·79–1·06
10·5 0·3 8 0·15 0·33 2·2 0·73 0·5–0·67 0·37–0·49
10·5 0·5 8 0·15 0·55 2·2 1·21 0·5–0·67 0·61–0·81
10·5 0·8 8 0·15 0·88 2·2 1·94 0·5–0·67 0·97–1·30
12 0·3 8 0·15 0·33 2·5 0·83 0·5–0·67 0·42–0·56
12 0·5 8 0·15 0·55 2·5 1·38 0·5–0·67 0·69–0·92
12 0·8 8 0·15 0·88 2·5 2·20 0·5–0·67 1·10–1·47
v,speedoftowingunit;n,empiricalfactorindepthofimprovementequation(lowervaluesofnforclay,highervaluesofnforgranularsoils,intermediatevaluesofn
formixedsoils);m,modulemass;h,maximummoduledropheight;D,depthofsoilcompactedduetogravitationalpotentialenergy;k,ratiooftheenergyimparted
tothegrounddividedbygravitationalpotentialenergy;r,reductionfactorfordeterminingDMI
towing speed than forclay soils. However, not all ground con- Acknowledgements
ditions cansustain atowingspeed of 12km/h for the8 tfour- The authors acknowledge the following final year civil engin-
sided impact roller; therefore, in the absence of site-specific eering students from the Universityof Adelaidewho contribu-
information, a median towing speed of 10·5km/h is rec- ted to the fieldwork referred to in this paper: Nicole Mentha,
ommendedforuseinTable6. Simon Pointon, Erfan Syamsuddin, Aidan Symons and
Penelope Wrightson. The authors are grateful to the instru-
mentation and laboratory staff at the University of Adelaide
7. Conclusions
who have provided invaluable support over the years and to
Thispaperexaminedimprovinggroundinsituandcompaction
Cathy Cates for her assistance with word processing. The
of soil in thick layers astheyare two distinctly different appli-
authors are alsograteful to Broons and HWE Mining person-
cations for RDC that, in the authors’ opinion, need to be
nelforaccesstoequipmentandthetestsite;withouttheirhelp
treated independently. Foratowing speed of 10·5 km/hfor the
and support, thework undertaken in the field trial featured in
8t four-sided impact roller, the EDI was estimated to be
thispaperwouldnothavebeenpossible.
0·73 m for clay soils (n=0·3) and 1·94 m for granular soils
(n=0·8). This highlights that soil type is the single most
important variable in quantifying the depth to which RDC
can improve soil. A relationship to evaluate EDI is presented REFERENCES
ASTM(2007)D698:Standardtestmethodsforlaboratorycompaction
as a function of the energy imparted to the ground by RDC,
characteristicsofsoilusingstandardeffort.ASTMInternational,
which is appropriate for determining the depths to which WestConshohocken,PA,USA.
groundcanbeimprovedinsitu.Forthefieldtrialpresentedin ASTM(2008)D5195:Standardtestmethodsfordensityofsoiland
this paper, an EDI of 2·0m was measured using buried EPCs rockin-placeatdepthsbelowsurfacebynuclearmethods.
ASTMInternational,WestConshohocken,PA,USA.
andcomplementaryinsitutesting.
ASTM(2009a)D6913:Standardtestmethodsforparticle-size
distribution(gradation)ofsoilsusingsieveanalysis.
A second relationship to determine DMI, is also introduced, ASTMInternational,WestConshohocken,PA,USA.
which is appropriate for determining the thickness of layers ASTM(2009b)D1557:Standardtestmethodsforlaboratory
thatcanbecompactedusingRDC,typicallyhalftotwothirds compactioncharacteristicsofsoilusingmodifiedeffort.
ASTMInternational,WestConshohocken,PA,USA.
of the EDI. For the field trial presented in this paper, a DMI
ASTM(2010a)D2216:Standardtestmethodsforlaboratory
of1·2mwasmeasuredusinginsitutesting.Theequationspre-
determinationofwater(moisture)contentofsoilandrockby
sented in this paper augment the relationship for dynamic mass.ASTMInternational,WestConshohocken,PA,USA.
compaction first proposed by Menard and Broise (1975). ASTM(2010b)D4318:Standardtestmethodsforliquidlimit,
In addition to soil type, module mass and drop height, the plasticlimit,andplasticityindexofsoils.ASTMInternational,
WestConshohocken,PA,USA.
equations presented also incorporate the effect of towing
AvalleDL(2004)Groundimprovementusingthe‘square’impact
speed. While the equations presented in this paper are rela-
roller–casestudies.InProceedingsofthe5thInternational
tively simple in nature, the proposed energy-based approach ConferenceonGroundImprovementTechniques,KualaLumpur,
yields estimations of the depths capable of being significantly Malaysia(FaisalA(ed.)).CI-Premier,Singapore,pp.101–108.
improved in situ and the layer thicknesses capable of being AvalleDL(2007)Trialsandvalidationofdeepcompactionusingthe
‘square’impactroller.InSymposiumAdvancesinEarthworks.
compacted by RDC, which are in broad agreement with the
AustralianGeomechanicsSociety,Sydney,Australia,pp.63–69.
findingsofthefieldtrialpresentedandtheresultsofpublished
AvalleDLandCarterJP(2005)Evaluatingtheimprovementfrom
case studies involving the 8t four-sided impact rollerover the impactrollingonsand.InProceedingsofthe6thInternational
pastfourdecades. ConferenceonGroundImprovementTechniques,Coimbra,Portugal
9
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license |
ADE | 100 Scott,JaksaandMitchell
interpreting instrumented rollerdata. This studyovercomes work done. The key aims of this study are to measure the
previouslimitationsbyattachingaccelerometerstoanEPC loading-induced stresses and displacements that soil par-
andburyingtheminhomogeneous fillmaterialtoquantify ticlesbeneaththegroundsurfaceexperience,andtoquantify
theloadinginducedstressandgrounddecelerationbeneath theworkdonefrommeasuredforce–displacementdata.
thegroundsurface,yetwithintheexpectedzoneofinfluence
ofRDC.
RESEARCHTESTSITE
Figure 1 shows a four-sided 8 t impact roller (1450 mm
squareand1300 mmwidemodule)thatwasusedatadedi-
Comparisonswithdynamiccompaction
cated research site located at Monarto Quarries, approxi-
Measuring the ground response of deep dynamic compac-
mately 60 km south-east of Adelaide, Australia. While
tion has been studied by Mayne & Jones (1983), who
conducting a full-scale trial that is not associated with a
attached an accelerometer to a 20·9 t pounder to monitor
client-funded project is expensive, a research focused trial
the deceleration on impact with the ground surface after
providedanabilitytocontrolanumberofvariablesthatcan
fallingadistanceof18·3 m;thedeceleration–timeresponse
oftenconcealthetrueeffectsofRDC.Significantly,natural
of the impact blow occurred over a duration of only
soil was excavated to a depth of 1·5 m and replaced with
0·05 s.Alsoofsignificanceisthemagnitudeofdecelerations
homogeneousfill;acrushedrockwithamaximumparticle
recordedwasintheorderof70–85g,andatrendofincreas-
size of 20 mm that was readily available and locally
ing magnitude with number of drops was observed. Clegg
producedatthequarry.Sixequalliftsof250 mmthickness
(1980) attached an accelerometer to a falling weight and
wereadopted;thematerialwasplacedusingaVolvoL150E
found that the peak deceleration of the weight on impact
loader, and was lightly compacted using a 60 kg vibrating
with the soil was directly related to the soil resistance,
plateandwheelrollingfromtheloader.Thefillmaterialwas
described as a combination of both soil stiffness and
classifiedasawell-gradedsandygravel(GW)inaccordance
shearing resistance. Chow et al. (1990) developed a theor-
with the Unified Soil Classification System. The fill was
etical framework that was based on matching deceleration
tested for homogeneity through the use of particle-size
measurementsofadynamiccompactionpounderimpacting
distribution and Proctorcompaction testing; the results are
thegroundusinganaccelerometerthatwasattachedtothe
giveninTable1.
pounder near the centre of gravity. The one-dimensional
model that was developed was similar to pile driving ana-
lyseswheretheimpactvelocitywasobtainedbyintegrating
EPCsandaccelerometers
measured decelerations. Yu (2004) double integrated the
acceleration–time response of a vertically falling plate to FieldtrialsundertakenbyAvalleetal.(2009)andScottetal.
generate the load–displacement relationship, which was (2016) using the four-sided impact roller haveshown that a
integrated to quantify the work done. Analysis of a load– module impacting the ground directly above embedded
instrumentationresultsinsignificantlyhighergrounddecel-
displacementresponseduetoimpactwasalsoundertakenby
erations being recorded, compared with when the module
Jha et al. (2012), who investigated energy dissipation to
strikes the ground off-set from embedded instrumentation.
quantifytheelasticenergythatwasrecoveredduringunload-
A limitation of burying equipment at discrete locations is
ingofmulti-phasecementitious materials.Theyplottedthe
load–displacement response for cementitious materials that it is not possible to capture the maximum ground
response from every impact. However, a key advantage of
subjected to nano-indentation and determined the area
this technique is that it does provide real-time data on
undertheloadingandunloading curvesandquantifiedthe
dynamic pressures and accelerationsthat are imparted into
thegroundthatothertestingmethodsareunabletodo.
A custom-built accelerometer cluster consisting of ±5g
and±16gaccelerometersintheZ-planetomeasurevertical
acceleration,and±5gaccelerometersintheX-andY-planes,
to measure tilt perpendicular to, and in the direction of
travel,respectively.Atotalof80passeswereundertaken.The
accelerometer cluster was attached to an EPC (230 mm
diameter and 6 mm thick) that was buried at a depth of
0·7 mbelowthegroundsurface,andconnectedtoabespoke
data-acquisition system and Labview software program
(refer Labview (2018)). The ability to capture an accurate
ground response using EPCs and accelerometers relies
heavilyon adopting a sufficiently high sampling frequency.
Giventhatdisplacementistobequantifiedfromthedouble
integrationofacceleration–timedata,asamplingfrequency
of 4 kHz (twice that adopted by Avalle et al., 2009) was
selected for this trial to ensure that the true peak pressure
Fig.1. Eighttonnefour-sidedimpactroller and ground deceleration could be accurately captured. As
Table1. Particle-sizedistribution,compactionandfieldmoisturetestresultsoffillmaterial
Material d : Gravel Sand Fines:% Standard Standard FMC:% Modified Modified
50
mm size:% size:% OMC:% MDD:kN/m3 OMC:% MDD:kN/m3
20mmcrushed 4·0 57 40 3 7·9 17·9 8·6 7·2 18·9
rock
d ,particlesizeatpercentfinerof50%;OMC,optimummoisturecontent;MDD,maximumdrydensity;FMC,fieldmoisturecontent.
50
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license |
ADE | Groundresponsetorollingdynamiccompaction 101
discussed by Thong et al. (2002), faster sampling rates can 0·05 s.Figure3illustratesthevertical(Z-)acceleration–time
improvetheaccuracyofintegration,buterrorscanincrease response for the same pass shown in Fig. 2, whereby a
withthedurationofthetimeintervaloverwhichintegration downward (negative) acceleration first occurs as the
isundertaken. soil is loaded. In response to loading, the soil resistance
ismobilised,whichresultsinanupwardaccelerationbefore
the acceleration trace dampens and returned to zero less
RESULTSANDDISCUSSION than 0·1 s after loading. Significantly, a peak deceleration
A single pass (54 summarised in Table 2) was selected out (negative acceleration) of 21g was measured before the
of the 80 passes undertaken for analysis as it featured a soil resistance was mobilised. Table 3 includes a summary
high peak pressure and the largest vertical deceleration of passes 1–10, as well as every fifth pass thereafter.
recorded.InFig.2themoduleimpactresultedinameasured As observed in Table 3, the magnitude of the peak down-
peak pressure of 1077 kPa at a depth of 0·7 m. It can be ward acceleration was typically greater than the peak
observed that the impulse pressure imparted to the ground upward acceleration, this trend was more defined for
wasloadedandunloadedoveradurationofapproximately impacts that generated large accelerations. Consequently, a
Table2. Summaryofpass54fortestdepthof0·7m
Pass δ : δ : W : W : W : Peakpressure: Δt:s Peak Peak
elastic plastic total elastic plastic
mm mm J J J kPa deceleration:g acceleration:g
54 4 5 254 36 218 1077 0·05 −21·0 6·3
δ ,reboundsettlement;δ ,permanentsettlement;W ,totalareaunderload–displacementcurve;W ,elasticworkdone;
elastic plastic total elastic
W plasticworkdone;Δt,durationofappliedload;peakdec.,peakdeceleration;peakacc.,peakacceleration.
plastic=
1200
1077
1000
800
600
400
200
0
4·40 4·41 4·42 4·43 4·44 4·45 4·46 4·47 4·48 4·49 4·50
Time: s
–200 Δt ≈ 0·05 s
Fig.2. Pressuredistributionattimeofmoduleimpact
10
6·3
5
0
4·40 4·41 4·42 4·43 4·44 4·45 4·46 4·47 4·48 4·49 4·50
–5 Time: s
–10
–15
–20 X
EPC Y
–21·0
Z
–25
Fig.3. Z-accelerationresponseattimeofmoduleimpact
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license
g
:noitarelecca-Z
aPk
:erusserP |
ADE | 102 Scott,JaksaandMitchell
shift in the baseline (zero) reading was undertaken that Figure4showsaplotofY-acceleration(inthedirectionof
enabled readings of −21g and +6·3g to be measured traveloftheroller)againsttime.Ofsignificanceinthisplotis
using a ±16g accelerometer (range of 32g). Consistent the larger magnitude of the positive (compared with
with the findings of Mayne & Jones (1983), an increased negative)Y-acceleration.Itcanbeinferredthatthedirection
number of passes generally resulted in larger accelerations of travel of the module influences the ground response, an
(and peak pressures) being recorded. However, the vari- expected result given the module drop is not solely in a
able location of the module landing on the ground vertical direction. Figure 5 shows a plot of X-acceleration
surface relative to buried instrumentation, analysed and (perpendicular to the direction of travel) with time. Both
discussed by Scott et al. (2016), was also a contributing positiveand negative accelerations areapproximatelyequal
factor that would explain why some passes (e.g. pass 54) suggestingthatthemodulelandingdirectlyoverthecentreof
yieldedmuchlargerpeakpressuresandverticalaccelerations the cell produces a relatively symmetrical response in the
thanothers. directionacrossthetestlane,thisisnotunexpectedgiventhe
Table3. Summaryofpassesfortestdepthof0·7m
Pass δ : δ : W :J W :J W :J Peakpressure: ImpulseΔt: Peakdec.: Peakacc.:
elastic plastic total elastic plastic
mm mm kPa s g g
1 2·0 0·5 13 9 4 230 0·07 −3·5 3·0
2 3 1 44 13 31 419 0·07 −5·5 3·8
3 3·5 0·5 35 25 10 371 0·08 −5·3 4·4
4 3 2 76 20 56 594 0·08 −4·6 2·5
5 6·5 0 108 53 55 656 0·07 −5·6 7·7
6 3 2 71 13 58 503 0·06 −11·6 5·2
7 3 2 64 20 44 550 0·08 −2·1 3·4
8 1 1 73 45 28 177 0·08 −1·3 0·6
9 2 1 22 6 16 258 0·05 −4·9 2·8
10 3 2 71 14 57 539 0·06 −8·5 3·9
15 3 2 56 15 41 490 0·08 −4·0 1·7
20 3 2 62 18 44 492 0·05 −9·6 4·8
25 2·5 1·5 35 14 21 324 0·06 −8·0 4·7
30 6 0·5 58 29 29 380 0·06 −10·5 9·6
35 2·5 1 22 7 15 272 0·05 −4·0 2·9
40 2 3 41 5 36 309 0·04 −6·6 4·4
45 2·5 0·5 12 4 8 166 0·05 −1·6 2·6
50 2 1 11 7 4 202 0·06 −1·8 1·7
55 3·5 2·5 98 24 74 680 0·05 −7·2 5·6
60 2·5 0·5 11 7 4 169 0·07 −2·4 2·5
65 3·5 3·5 177 14 163 873 0·05 −13·2 5·4
70 4 1·5 60 34 26 557 0·07 −4·9 3·8
75 1·5 6 136 18 118 731 0·07 −9·2 4·5
80 7·5 0·5 249 59 190 1115 0·05 −11·2 8·0
δ ,reboundsettlement;δ ,permanentsettlement;W ,totalareaunderload–displacementcurve;W ,elasticworkdone;
elastic plastic total elastic
W ,plasticworkdone;Δt,durationofappliedload;Peakdec.,peakdeceleration;Peakacc.,peakacceleration,peakvaluesinbold.
plastic
Y-acceleration: g
–1·0 –0·8 –0·6 –0·4 –0·2 0 0·2 0·4 0·6 0·8 1·0
4·40
4·42
4·44
0·71
4·46
4·48
4·50
–0·27
4·52
4·54
4·56
X
4·58 EPC Y
Time: s Z
4·60
Fig.4. Y-accelerationresponseattimeofmoduleimpact
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license |
ADE | Groundresponsetorollingdynamiccompaction 103
X-acceleration: g
–1·0 –0·8 –0·6 –0·4 –0·2 0 0·2 0·4 0·6 0·8 1·0
4·40
4·42
0·31
4·44
–0·29 4·46
4·48
4·50
4·52
4·54
4·56
X
4·58
EPC Y
Time: s Z
4·60
Fig.5. X-accelerationresponseattimeofmoduleimpact
100 0·004
Z-acceleration Z-displacement
50 0·002
Time: s
0 0
4·40 4·41 4·42 4·43 4·44 4·45 4·46 4·47 4·48 4·49 4·50
–50 –0·002
–100 δ plastic ≈ 5 mm –0·004
–150 –0·006
–200 X –0·008
EPC Y
Z
–250 –0·010
Fig.6. Z-accelerationandZ-displacementagainsttime
moduleonlyhasalimitedabilitytomovelaterallywithinthe double integration of the acceleration–time response. In
trailerframe. Fig. 7 the portion of the curve between points A and B
Figure 6 shows the variation of Z-acceleration and representstheloadingofthesoil.Theunloadingportionof
Z-displacementofthesoilwithtimeinresponsetoasingle the curve is shown between points B and C. The distance
module impact, whereby displacement was calculated from between points A and C provides a measure of the perma-
double integration of the acceleration–time response. From nent deformation of the soil. For a perfectly elastic soil
Fig.6,itisevidentthatapproximately9 mmtotaldisplace- response with no hysteresis, AB and BC would be coinci-
ment occurred due to loading; however, on unloading, the dental.AreaABCyieldstheplasticworkdoneandthearea
permanentdisplacementduetothesingleimpactwas5 mm. CBD represents the elastic work that has been recovered
The same impact blow is illustrated in Fig. 7, which shows during unloading. The total work done comprises both
theloadingandunloadingresponseofthesoilduetoasingle recoverable(elastic)andpermanent (plastic)components.
pass of the impact roller at a measured depth of 0·7 m Figure 8 showsthe force–displacement response forcon-
beneaththegroundsurface.Forceisdeterminedbyadopting secutivemoduleimpacts(passes1–10inclusive,summarised
thepeakpressureatthetimeofimpactandmultiplyingitby inTable3).Ascanbeobserved,thereisalargevariationin
the plan area of the EPC. Displacement is evaluated from theshapeandmagnitudesoftheforce–displacementcurves
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license
2s/m
:noitarelecca-Z
Z-Displacement:
m |
ADE | 104 Scott,JaksaandMitchell
50
45 B
40
35
30
25
20
15
W
plastic
10
5
A C W elastic D
0
0 1 2 3 4 5 6 7 8 9 10
Displacement: mm
Fig.7. Force–displacementcurveforasinglepass
30
5
4
25
7 10
6
20
2
3
15
9
1
10
8
5
0
0 2 4 6 8 10 12 14 16
Displacement: mm
Fig.8. Force–displacementcurvesforconsecutivepasses(1–10)
for individual passes. Pass 1 is close to an elastic impact of passes required to significantly improve ground to meet
whereminimalworkisdoneonthesoil.Theoppositeistrue acertainspecifiedcriterion.Whilethenumberofpasses(80)
for pass 10, which features a much larger area under the undertaken in this study was greater than what would
force–displacementcurve. economically be undertaken in practice, the results from
buriedinstrumentationindicatethat0·7 miswellwithinthe
depth range that can be significantly improved by RDC.
Quantifying the dynamic behaviourof the soil beneath the
CONCLUSIONS
ground surface in real-time emphasises that the uneven
To minimise soil variability, this study has captured the
module geometry results in some passes imparting much
change in vertical stress due to RDC at a depth of 0·7 m
greater pressureto the ground than others, this being a key
beneath the surface using an EPC buried in a 1·5 m thick
reason why many passes are needed to ensure adequate
layerof homogeneous sandygravel. The maximum change
coverageofasite.
in vertical stress recorded over the 80 passes undertaken
was approximately 1100 kPa. During a typical module
impact, the loading and unloading of the soil occurred
overadurationofroughly0·05 s.Theaccelerationresponse ACKNOWLEDGEMENTS
of a single module impact was also measured in three The authors acknowledge the following final year civil
orthogonal directions at 0·7 m depth, with the vertical engineering students at the University of Adelaide who
accelerations dominant. In project applications, there is assistedwiththefieldworkreferredtointhispaper:Stefan
typicallyatrade-offbetweenlayerthicknessandthenumber Chenoweth, Jordan Colbert, Julianne Sawand Ross Vince.
Downloaded by [] on [17/10/19]. Published with permission by the ICE under the CC-BY license
Nk
:ecroF
Nk
:ecroF |
ADE | Abstract
Water supply and distribution systems are an integral part of our society and can incur significant costs
in their construction and operation. Many different optimization techniques have been applied to both
the design and operation of traditional potable systems, which generally receive water from natural
water bodies. As climate change and increasing populations prompt concerns of water security, in
addition to natural harvested water supplies, alternative sources such as harvested stormwater,
recycled wastewater and desalination are becoming more commonly used for both potable and non-
potable supply. These systems have not been researched as extensively, particularly their operation.
This thesis examines the optimisation of pumping operations in water supply and distribution systems
that can include conventional potable systems as well as systems that use alternative water sources.
The major contributions of this research are presented in three publications. Firstly, a single-objective
optimisation model was applied to potable water distribution systems, both hypothetical and real, for
different types of pump operating regimes using the EPANET toolkit to alter rule-based controls. While
minimizing pump energy costs was the primary objective, minimization of greenhouse gas emissions
was also explored, including the variation of greenhouse gas emission factors for different electrical
energy sources. Time-based scheduling operating strategies were found to perform better than the
other operating regimes, and significant cost savings were achieved for the real-life system compared to
its current operation.
In the second paper, a framework for the optimization of water supply and distribution systems that use
alternative water sources is presented, along with a detailed discussion of the components and key
variables. The framework connects the potential decision variables, both design and operational, the
physical components of the water system to be modelled, the simulation of each potential system
configuration and evaluation against objectives and constraints, and relevant policies from regulating
bodies. These all exist within an optimization algorithm structure, and sensitivity analysis of uncertain
variables is also discussed. Two case study systems are used to illustrate how the framework would be
applied to minimize the cost of water system operations.
The final paper applies multi-objective optimisation techniques to a harvested stormwater case study
system and also covers an extensive sensitivity analysis of the inputs to the system. This system has
distinct winter (harvesting) and summer (irrigation) operational seasons; for the winter operation,
operating rules were optimized to minimize the cost of pumping into an aquifer and to maximize the
volume harvested, considering restrictions on the aquifer injection rate and pressure; during summer,
irrigation scheduling was optimized to minimize pumping costs, considering the requirements for
irrigation rates and amounts at various public parks and green area reserves. Results from both the
optimisation and sensitivity analysis found operational cost savings if new pumps are installed, wider
trigger levels are used, and certain reserves are irrigated together.
This research has produced significant overall contributions to the body of knowledge. Methodologies
have been developed for optimisation of potable and alternative water sources systems, highlighting
important considerations and generalizable results. For three real-life case study systems, operating
strategies and infrastructure changes have been suggested to provide significant savings in the cost of
pumping operations.
x |
ADE | Chapter 1 Introduction
1.1 Research Background
Water supply and distribution systems are vital parts of today’s society, ensuring the health of our
communities and providing commercial, industrial and recreational benefits. These systems can have
high construction and operating costs, as well as associated greenhouse gas (GHG) emissions and have
long been the focus of research to make them more efficient, lower cost, and more reliable, among other
objectives. Climate change is a major concern for society as a whole, and also for water resources
managers. Different regions of the world will experience the effects of climate change in different ways;
some areas will experience drying, while others will be wetter and the variability of rainfall is likely to
increase. Climate change will also affect how rainfall is translated into runoff, as climate conditions affect
the ability of soil and plants to intercept and retain water. This has major implications for how we obtain
our water supply, as many regions around the world source their water from natural catchment runoff. An
increasing population into the future will also put a strain on water resources. In light of this, alternative
water sources are increasingly being sought out by water system managers to provide security of water
supply into the future.
Some alternative sources of water, such as groundwater and imported catchment water have historically
been used in conjunction with natural catchment waters. Other sources, such as harvested stormwater,
treated wastewater, desalination, and aquifer storage and recovery (ASR) have gained popularity more
recently. Groundwater from aquifers may be used for various applications, depending on the quality of
the water. In some cases, it may already be at potable standard, or able to be further treated with little
cost to obtain potable standard, and therefore be used in mains distribution systems. If it is not of potable
standard, it is often used for irrigation of private gardens and public parks and reserves, especially when
water restrictions are put in place to limit outdoor irrigation with mains water. Imported water is often used
in areas with low local rainfall, obtaining water from other areas with higher rainfall or significant water
bodies through long pipelines or canals. Harvesting of urban stormwater runoff is often applied at
community scales to provide water for irrigation of public spaces. It can provide other benefits such as
reducing urban runoff and creating amenity in public recreation areas. Desalination plants, while energy
intensive and expensive to run, provide a climate independent source of water, and as such is a popular
choice for regions prone to long or intense droughts. Recycled wastewater is another source that is also
climate independent and is often used for non-potable supply, such as large scale irrigation or industrial
use. Advances in treatment technologies have allowed potable standard water to be produced from
wastewater, however, public perception regarding the acceptability of usage still lags behind. Stormwater
harvesting and wastewater recycling systems are sometimes combined with ASR, allowing water to be
stored for long periods of time in an underground aquifer and utilized when needed (without the need for
large storage tanks or above-ground reservoirs that would have large construction costs and reduce
amenity of public spaces). On a household scale, rainwater tanks are used to collect water from roofs
generally for outdoor irrigation, however, this water may also be used indoors and for drinking. Greywater
recycling systems are also gaining popularity, typically re-purposing water from showers, taps and
washing machines for outdoor irrigation.
Uptake of alternative water source systems has been restricted by public and industry perception, cost,
and development of appropriate technologies. While alternative sources can be, or are treated to potable
quality, there is a perception that they are not suitable for drinking or human contact. The public often do
not want to use alternative sources such as stormwater and recycled wastewater where there is the
potential for human contact, which has restricted their application. As many systems using alternative
sources are on small, decentralized (local) scales, technology to capture, treat and store water may not
1 |
ADE | Introduction
be available at the appropriate capacity or at a reasonable cost. The design and operation of these smaller
scale systems may not necessarily be handled by people with the required expertise, and as a result the
system will not perform as well as desired. Natural catchment water is a relatively low cost source, as the
infrastructure to capture the water is usually already in place, the main ongoing cost is the treatment of
the water. Developing alternative water source systems requires more capital infrastructure costs, and
may also require higher levels of treatment or transportation over long distances, therefore increasing
their ongoing costs compared to existing resources.
Energy use is one of the major contributors to ongoing costs in water distribution systems (WDSs).
Reducing their energy use starts in the design phase, investing more in capital infrastructure may allow
the system to operate with less energy requirements and therefore reduce ongoing costs. There is usually
a trade-off between capital and ongoing pumping costs that should be explored to find the best
compromise for a particular system. For existing systems, energy efficiency can be improved using
strategies such as leak identification and repair or system maintenance as well as by altering the pump
operating rules of the system. Variable speed pumps (VSPs) can also be used to adjust the pump
operating points for different system conditions and save energy by reducing pumping heads and flows.
In systems where excess pressure energy occurs, it may be recovered using mini-hydro systems or
pumps and turbines. Pump operating strategies can broadly be split into trigger levels (based on the
amount of water or level in a storage) and scheduling (based on the time of day or week). Electricity tariff
periods should be considered when optimizing pump operating rules, and different rules may be required
for different seasonal conditions.
While engineering judgement can be used to guide the design and operation WDSs successfully, there is
often a large number of decisions to be made and multiple objectives. Formal optimization algorithms are
very useful in order to efficiently find solutions that will improve the performance of the system with regard
to the objectives. They do not necessarily need to analyse all possible solutions to find the optimal
solution(s). When multiple objectives exist, care needs to be taken when determining the objective
function(s). Multiple objectives can often be combined into one function, however, this requires the
normalization of objective values and the relative importance of each objective needs to be decided upon.
There are many multi-objective optimization algorithms available, that are able to deal with each objective
function separately, allowing them to retain more meaning. Engineering judgement should always be used
in conjunction with optimization, as it can help to limit the search space of the problem and ensure the
optimal solutions found are reasonable. Simulation of the system prior to optimization is very important
as it provides an understanding of how the system works and helps these engineering judgements to be
made. Genetic algorithms (GAs) are a robust and efficient optimization method that have been used
extensively for the design and operation of WDSs. They are a population based technique, which means
they evaluate multiple solutions at once and use operators based on natural selection principles to
gradually improve the performance of the population through successive generations. Given the
complexity and cost constraints of alternative water source systems, optimization methods such as GAs
are very useful to improve their performance and make them more cost comparable to traditional WDSs.
1.2 Research Objectives
The overall aim of this research is to develop and apply methodologies for optimizing complex pumping
operations to systems that use alternative water sources; this is split into six objectives:
Objective 1. To develop a framework to optimize alternative water system pump operations for multiple
objectives including minimizing cost and maximizing volume harvested.
2 |
ADE | Introduction
Objective 2. To apply the use of new rule-based controls in a modified EPANET2 programmer’s toolkit
to optimize complex pump operational strategies using a combination of trigger levels and
scheduling, and variable trigger levels.
Objective 3. To optimize pumping operations and irrigation scheduling for short time horizons for
systems using harvested stormwater with aquifer storage and recovery and multiple
pumping stations.
Objective 4. To demonstrate the importance of performing detailed simulation analysis of water systems
in order to better understand the system and to inform optimization of the system.
Objective 5. To analyse the sensitivity of optimal pump operations to changes in streamflow (system
inflow) and system design in a stormwater harvesting system.
Objective 6. To minimize GHG emissions from pump operations where operational characteristics are
considered as decision variables and characterize trade-offs between optimal cost and
optimal GHG solutions for these problems.
As shown in Figure 1.1, the six objectives are connected and each of the Chapters in the main body will
contribute to multiple objectives. The development of a framework in Objective 1 will inform the execution
of Objectives 3 and 6. Rule-based controls in a modified EPANET2, which are specifically included within
Objective 2, will also be used in Objectives 3 and 6. The detailed analyses in Objectives 4 and 5 will
inform the optimization of a harvested stormwater system in Objective 3. Objectives 2 and 6 represent a
gap in the current research on optimization of pump operations in potable WDSs and are investigated in
the Chapter 4 for two potable WDS case studies. Chapter 5 investigates Objective 1, and how the current
methodologies used for potable WDSs need to be altered to take into account additional complexity and
processes that come with the use of alternative water sources. It also discusses variables that should be
taken into account in sensitivity analyses of water systems, such as in Objective 5. Objectives 3, 4 and 5
are addressed in Chapter 6, which details the analysis and optimization of pumping operations in a
harvested stormwater and ASR case study.
1 2
Rule-based controls
Framework
in a modified EPANET2
Chapter 5
Publication 2
3 6
Optimize harvested Pumping cost and Chapter 4
stormwater operations GHG trade-off Publication 1
4 5
Pre-optimization
Sensitivity analysis
analysis by simulation
Chapter 6
Publication 3
Figure 1.1: Connections between the six objectives and chapters in this thesis
3 |
ADE | Introduction
1.3 Thesis Outline
This thesis is presented as a collection of three journal publications that were developed along with the
research undertaken and is arranged in seven chapters. Chapter 2 presents a detailed review of the
relevant literature on the topics of pumping operations, alternative water sources and genetic algorithm
optimization. The three publications that make up this work are summarised in Chapter 3, which
demonstrates how the publications are linked to each other and to the research objectives identified in
Section 1.2.
Chapter 4 presents the first publication (Blinco et al. 2016a): ‘Comparison of pumping regimes for water
distribution system to minimize cost and greenhouse gas emission,’ published in the Journal of Water
Resources Planning and Management. In this paper, five different types of pump regimes were explored;
lower and upper trigger levels, reduced upper trigger level, combined trigger levels and scheduling,
variable trigger levels, and variable speed pump (VSP) scheduling (Objective 2). These regimes were
optimized and compared for two potable case study networks, considering objectives of minimizing pump
energy costs and minimizing GHG emissions from pumping (Objective 6).
The second publication (Blinco et al. 2017a) is in Chapter 5: ‘Framework for the optimization of operation
and design of systems with different alternative water sources,’ published in Earth Perspectives. This
paper presents a methodology for optimizing water supply and distribution systems that use alternative
water sources such as harvested stormwater, imported water (from adjacent catchments), groundwater
and desalination (Objective 1). The framework details the different design and operational options, the
water and electrical energy infrastructure, the relevant government policies, the simulation model and
evaluation options and how these components fit within and optimization algorithm. Variables that may
be considered in sensitivity analyses of water systems are also discussed (Objective 5) and two case
studies are used to demonstrate the application of the framework.
Chapter 6 contains the final publication (Blinco et al. 2017c): ‘Optimization of pumping costs and
harvested volume for a stormwater harvesting system,’ submitted to the Journal of Water Resources
Planning and Management. This paper demonstrates the application of the framework methodology from
the second publication, and the pumping operations optimization from the first paper to a harvested
stormwater system (Objective 3). The first part of the paper shows a detailed analysis of the current
operation of the system and possible operation under different design scenarios (Objective 5).
Optimization of the system is then presented and the importance of both the pre-analysis and optimization
procedures is discussed (Objective 4).
The main conclusions and contributions of this research are presented in Chapter 7. This chapter also
summarizes the limitations of the research and suggested future directions in this study area.
4 |
ADE | Chapter 2 Literature Review
2.1 Pumping Operations
The operations stage of a WDS is a significant contributor to life-cycle energy use (Stokes and Horvath
2005) and therefore often represents a significant cost to water utilities (Boulos et al. 2001). Optimizing
how WDSs operate, particularly in terms of pumping controls, can therefore have a significant impact on
reducing cost and energy use for water system managers. Other strategies for recovering or reducing
energy use in WDSs include energy dissipation by mini-hydro systems or pumps as turbines (Carravetta
et al. 2013b, Fecarotta et al. 2015), leak reduction (Giustolisi et al. 2013) and system maintenance or
repairs. Cabrera et al. (2016) highlight the importance of examining ‘topographical energy’, that is excess
pressure at nodes of low elevation, in a network. Where large amounts of topographical energy exist,
pumps as turbines can be used to recover some, or pressure reducing valves can be installed to reduce
leaks. As well as cost, there are other objectives that may improve the operation of WDSs, such as water
quality (Stokes et al. 2012a), pump switches or maintenance cost (Lansey and Awumah 1994, López-
Ibáñez et al. 2005), system effectiveness (Carravetta et al. 2013a), and resilience (Prasad and Park
2003). The design of the system also has a significant impact on the ongoing energy use and there is
often a trade-off between initial construction costs and ongoing operational costs. Networks with smaller
diameter pipes have increased friction losses compared to those with larger diameter pipes, and hence
require more energy during pumping operations (Wu et al. 2011). This means that while smaller diameter
networks are generally less expensive to construct, they are more expensive to operate than larger
diameter networks and there will be a different compromise between capital and operational costs for
different systems. For existing system rehabilitation, installing newer, smoother pipes, or replacing pumps
with more efficient ones, usually may incur a significant capital cost, however, these actions can reduce
ongoing operational costs (Fernández Garciá et al. 2016). Elevated storages in a network used to store
water judiciously, can be used to reduce the amount of pumping in peak periods, therefore reducing
energy costs (Jin et al. 2015). Where energy sources with higher air pollutant emission rates are used as
top-up during times of peak electricity demands, the environmental impact of pumping can also be
reduced (Jin et al. 2015). An initial step to reducing the energy use of a WDS is to conduct an energy
assessment (Cabrera et al. 2010, 2015) to determine which parts of the system should be the focus for
removing energy inefficiencies. The research presented in this thesis is focussed on optimizing energy
cost of pumping operations.
There are two main types of pumping controls; trigger levels, which turn pumps on or off depending on
the level or volume in a storage, and scheduling, which requires pumps to be on or off at particular times
of the day. Both have been investigated extensively by optimization to reduce costs of WDS operation.
An important result is the benefit of pumping only in off-peak (lower cost) electricity tariff periods,
investigated in Mäckle et al. (1995) for pump scheduling and Kazantzis et al. (2002) for combined trigger
levels and pump scheduling. Both of these studies found that optimal solutions occurred when tanks were
full at the start of the peak tariff period, and at their minimum allowable level at the end of the peak tariff
period (Figure 2.1). This meant that the minimum possible amount of pumping would occur at the
expensive tariff rate, and the maximum possible amount of pumping at the lower cost tariff rate. For
systems with multiple pumps, the most efficient pumps should be used during the peak (expensive)
electricity tariff period, and the least efficient during the off-peak period (Mäckle et al. 1995). Two type of
alterations to typical lower and upper trigger levels were examined in Kazantzis et al. (2002); adding a
scheduled pump start and pump stop, or using a reduced upper trigger level. A pump stop can be
scheduled before the end of the peak period, to ensure the water level in the tank is at the minimum
allowable level at the end of this period. Likewise, a scheduled pump start before the end of the off-peak
period, can ensure the water level is at the maximum allowable level for the start of the peak period. A
5 |
ADE | Literature Review
reduced upper trigger level applied over the peak tariff period will limit the static head of the system, and
therefore less energy will be required for pumping. At a specified switch time during the off-peak period,
the reduced upper trigger level will be removed so that the tank can fill before the start of the peak period.
The solution presented in Kazantzis et al. (2002) optimized the reduced upper trigger level, a scheduled
pump stop, and the switch time for the reduced level. Lower and upper trigger levels were used in the
solution, however, they were not optimized.
Maximum allowable level
Tank water level
le
v
e
l
k
n
a
T
Minimum allowable level
Maximum tank level Minimum tank level Maximum tank level at
at start of peak period at end of peak period end of off-peak period
Peak tariff period Off-peak tariff period
Time
Note: it is difficult to achieve these tank level criteria using only a lower and upper trigger level
(adapted from Kazantziset al. 2002)
Figure 2.1: Example of tank water level with efficient pumping in WDSs (adapted from Kazantzis et al. (2002))
While peak and off-peak tariffs are an important consideration for cost minimization, in order to reduce
GHG emissions, it may be better to pump steadily throughout the day with a VSP to reduce the velocity
of flow in the pipe and hence reduce the friction loss (Simpson 2009). Lingireddy and Wood (1998) and
Wu et al. (2011) have demonstrated the benefits of using VSPs to reduce both energy use and GHG
emissions in WDSs. They are particularly effective in smaller diameter networks with high friction losses,
as VSPs run at reduced flows, they can reduce the friction losses through the system (Wu et al. 2011).
The relative speed of VSPs may be a decision variable in an optimization formulation. In systems
controlled by trigger levels, the VSP speed at discrete time intervals during the day could be optimized,
which would be overridden by the trigger levels if they require the pump(s) to be off. The inclusion of VSP
decision variables in pump scheduling optimization depends on the form of the schedule. Pump
scheduling may be structured in two different ways; firstly using a discrete on or off (1 or 0, or VSP relative
speeds) at set time intervals (say every hour in a 24 hour simulation), or represented as continuous values
with set times (for example, 8:15am or 12:35pm) to turn pumps on or off. Continuous representation is
more flexible, however, can produce a high proportion of infeasible solutions depending on the coding of
the optimization algorithm (Sadatiyan Abkenar et al. 2015) and would require additional decision variables
to set the speed of VSPs.
Many studies into pump operations of WDSs use EPANET2 hydraulic simulation software to determine
energy use and cost of the systems (for example Kazantzis et al. 2002, López-Ibañez et al. 2005, and
Fernández Garciá et al. 2014). Gómez et al. 2016 examine the limitations and errors in EPANET with
regards to energy, which should be considered and addressed if needed when using the software. Three
major issues and four minor issues were raised. The first major issue was the error in calculating the
efficiency of VSPs operating at a reduced speed and this research utilized code to correct this error
(Marchi and Simpson 2013). The second major error is that ‘natural’ energy (from elevated tanks and
storages) is ignored, which may be important for performing energy audits (as in Cabrera et al. 2010) or
6 |
ADE | Literature Review
when considering different system layouts. This thesis is focussed on operations of existing systems (no
layout changes are considered) and minimizing electrical energy use, and thus this limitation is not
relevant to the current work. The final major issue raised is that the energy use and costs presented in
the EPANET2 interface are scaled to a 24 hour time period, even if the simulation is run for a different
length of time. When connected to an optimization algorithm, the energy cost can be calculated outside
of EPANET2 based on the energy use in each time step, thus avoiding the problem.
Two of the minor problems relate to the specification of electricity price tariffs, in particular for systems
with multiple pumping stations. Tariff patterns can be specified for each pump individually in order to take
into account changes to electricity prices over the simulation period (typically this represents daily or
weekly peak and off-peak tariffs). The peak power demand charge, however, is usually set for the whole
system, not each pump, which may be limiting. If a peak power demand charge applies to only some
pumps, or differs across pumps, external code (outside of EPANET2) may need to be used to accurately
compute the cost. The energy efficiency of variable speed drives (VSDs) and electric motors was another
issue raised, as EPANET2 considers only the pump efficiency. Both the motor and VSD efficiencies are
typically much higher than pump efficiencies, and if the pump speed is reduced to no less than 75% of
full speed, the pump efficiency needs to be altered (Sârbu and Borza 1998). If no VSPs are used, the
motor and VSD efficiencies do not change (whereas the pump efficiency may change with the pump
operating point), and as such will be the same for all operating strategies. While the energy costs
computed will not take into account motor and VSD efficiencies, they can still be compared between
different operating strategies as the effect of these other efficiencies would be the same for each strategy.
The final minor issue raised was the energy intensity (the energy used per volume), which is calculated
based on the volume supplied by pumps rather than the volume received by consumers (therefore
ignoring leaks). For systems with leaks, external code (to EPANET2) could again be used to work around
this problem.
A recent advance for EPANET2 is the additional capability of the programmer’s toolkit developed by
Marchi et al. (2016b) to allow rule-based controls to be optimized. Previously, only simple controls (with
only one condition) and pump scheduling could be optimized through EPANET2. Optimization of rule-
based controls (as implemented by Marchi et al. (2016b)) provides much greater flexibility and complexity
to be considered in pump operations optimization. Rule-based controls in EPANET2 are made up of many
different components, including logical operators, EPANET2 objects (tanks, pipes and so on) and their
identifying indices, hydraulic and system variables (for example pressure, flow, clock-time), relational
operators, status (open or closed pipes, valves or pumps) and values of the variables. Using the new
EPANET2 modified toolkit from Marchi et al. (2016b), each of these components can be optimized
individually, or the entire rule can be optimized as a whole.
In WDS simulation and optimization, it is often assumed that water is available in an upstream storage
reservoir. This separates the distribution system from the supply system, and does not consider
uncertainty in supply. The main source of uncertainty for WDSs is therefore in the consumer demands,
which naturally fluctuate daily, weekly and seasonally, and will also vary into the future with population
and climate change. Most studies incorporate a daily diurnal variation in water demands, however,
seasonal variation is also an important consideration. Paschke et al. (2001) optimized tank trigger levels
considering different water demands in different seasons. During summer, when demands are higher, the
optimal trigger levels kept the water level higher in the tank, whilst during winter, the water level was
allowed to be lower in the tank, as demands were reduced. Basupi and Kapelan (2015) used Monte Carlo
simulation to find optimal WDS design and operation that was flexible to future changes in demand. They
assumed that the demand follows a normal distribution, with the mean and standard deviation increasing
7 |
ADE | Literature Review
over time to represent greater future uncertainty. Stochastic programming was used by Goryashko and
Nemirovski (2014) to determine optimal robust pump schedules; that is, operations that are feasible for
all demand realisations. In their methodology, complex systems with non-linear hydraulics need to be
reduced down to equivalent linear systems as they used linear programming for optimization. Eck et al.
(2015) examined how estimates of demand mean and covariance can be produced from smart meter
data, and then used to develop demand scenarios for robust valve operation optimization. They found
that incorporating only a small number of scenarios could give significant improvement in pressure
constraint violation with little cost increase. Marques et al. (2015) used a ‘real-options’ approach to
consider multiple future demands with two objectives; the first was the combination of economic costs
and GHG emissions (using a carbon price), and the second was the level of service. For their case study,
the ‘real-options’ method considered the probability of different possible WDS adaptations at three stages
over a 60 year horizon through a decision-tree structure.
2.2 Alternative Water Sources
Water is increasingly being seen as a fundamental and finite resource (Bogardi et al. 2012) and alternative
water sources are being used to supplement potable demand as climate change and population growth
highlight water security issues (Fielding et al. 2015). Decentralised harvested stormwater systems (often
managed by local councils in Australia) and household greywater recycling systems are popular for
supplying non-potable demands such as household gardens and public green spaces (Naylor et al. 2012).
At household scales, installation of rainwater tanks is increasing in popularity (Campisano and Modic
2012), which reduces consumption of water from utilities and decreases stormwater runoff from residential
areas. The millennium drought prompted several Australian cities to construct desalination plants (King
et al. 2012), providing a climate-independent source of water. Use of desalination is also increasing in
other areas of the world, however is not always the most cost effective or environmentally sustainable
source of water (Miller et al. 2015, Becker et al. 2010). Recycling of wastewater and greywater on
community and regional scales is also gaining popularity, often for non-potable applications (Muga and
Mihelcic 2008, Oron et al. 2014), however in some cases it may also be used for indirect potable supply
(Rodriguez et al. 2009). Recycling wastewater for re-use at the same site is becoming common,
particularly in industrial settings (Mariano-Romaro et al. 2007). Imported water refers to water transported
through pipe or canal systems from different regions and is already used in many major cities, for example,
Adelaide (from the Murray River) and Los Angeles and San Diego (from the Colorado River). This typically
requires a lot of energy even in well-designed or optimized systems, because of the distance the water
must travel and the height it needs to be lifted (Water in the West 2013).
An alternative strategy to supplementing potable supplies with other water sources is demand
management to reduce per capita demand (for example, Freidman et al. 2014). Such strategies should
be considered under future climate change and population growth (Dawadi and Ahmed 2013). This can
take on forms such as mandated outdoor irrigation times, water efficiency standards for shower heads,
taps, toilets and appliances, and awareness campaigns to encourage the public to use less water
(Berhanu et al. 2016). Smart metering, which is becoming more commonly used by water utilities, can
provide information for demand management, such as data for early leak detection and demand pattern
classification and forecasting (McKenna et al. 2014). Each of these alternative sources, along with
demand management strategies, play a role in delivering water security to towns and cities around the
world. Communities also value other benefits of alternative water sources, for example, improved
hydraulic function and water quality from stormwater schemes (Londoño Cadavid and Ando 2013).
Negative public perception can come from a low awareness or understanding of associated risks (Hwang
et al. 2006) and different types of sources will have different levels of acceptance by the public (Feilding
et al. 2015). One of the main barriers to uptake of alternative source systems from a water system
8 |
ADE | Literature Review
manager’s perspective is the cost of running and maintaining the system (Dobbie and Brown 2012, West
et al. 2016).
The inclusion of alternative sources in water supply system increases the complexity of system simulation
and the corresponding optimization problem (Paton et al. 2014). Marchi et al. (2016a) optimized the
design of a harvested stormwater system, taking into account climate change and externalities such as
reduced runoff to receiving water bodies and reduced urban stream flows. They highlighted the need to
consider the supply and distribution sides of the system together, the use of longer simulation times and
the inclusion of rainfall and evaporation scenarios as factors that increased the simulation complexity
compared to traditional WDSs. Optimization of alternative water source systems often considers
objectives and constraints other than just construction or ongoing costs. In groundwater systems, land
subsidence is an important consideration and can be reduced by extracting water intermittently (Wang et
al. 2009). Water quality may need to be considered, such as in Labadie et al. (2012), which optimized
releases from multiple stormwater reservoirs to reduce pollutant loadings on downstream waters. When
alternative water sources are used to supplement potable supply, the amount of water than can be
harvested from the system is a key variable. It may be the single objective of an optimization problem (for
example, Eusuff and Lansey 2004), or combined in a multi-objective optimization with design or
operational costs and other objectives (for example, Karamouz et al. 2007, McArdle et al. 2011, di Matteo
et al. 2016). Tsai et al. (2009) optimized pump schedules in an integrated surface and groundwater system
for six objectives (combined into one weighted objective function); minimum pump energy use, minimum
pressure violation, minimum tank residence time, minimum tank level deviation, minimum weekly
drawdown and maximum tank reliability. Through altering the weightings of the different objectives, they
found that some of the objectives were interrelated and some could act as surrogates for others, with
energy use and pressure violation being the most important. Factors such as pressure violation and tank
level balancing are often included in optimization problems as constraints, however they can also be
formulated as objectives.
Sustainability is often a key concern in alternative water source systems and can be evaluated using a
‘triple bottom line’ of economic, environmental and social criteria. Kang and Lansey (2012) optimized life-
cycle cost (economic), GHG emission (environmental) and system reliability (social) of a dual-pipe
network using recycled wastewater for non-potable supply. In comparison to single-pipe systems, the
dual-pipe systems were more expensive, however they performed better in terms of the environmental
and social criteria. McArdle et al. (2011) also considered three objectives; minimizing present-worth or
capital and ongoing costs (economic), maximising the amount of water harvested from a stormwater
scheme (environmental benefits to urban water system and increased water security), and minimizing the
size of a storage reservoir in a public park (therefore minimizing the impact on the social amenity of the
park).
Due to the complexity of WDS simulation and optimization, and the additional considerations for
alternative water sources, many different frameworks, methodologies and decision-support tools have
been developed. Stokes et al. (2014) presented a framework for the design and operation of WDSs using
traditional water sources. The focus of this framework was the water-energy nexus, with different energy
sources and GHG emissions factors included for consideration, and cost and GHG emission objectives.
There was no consideration of the supply side of the WDS or alternative water sources. A framework by
Ashbolt et al. (2014) can be used to optimize operating plans for water systems using surface water,
groundwater, desalination, recycled wastewater and imported water. Multiple objectives are incorporated
by weighting their importance and multiple replicates of inflows can be used for uncertainty analysis. The
design of the system is not included in the decision variables and the operations consider the levels in
9 |
ADE | Literature Review
main reservoirs that trigger different water sources to be used, not the operation of pumps and smaller
storages within the individual water source systems. Harvested stormwater systems have not often been
included in these frameworks, however, the methodology in Marchi et al. (2016a) optimizes the design of
ASR stormwater systems with consideration of future climate scenarios. Externalities are included in the
analysis, such as reduced volume of stormwater to treat before discharge, reduced peak flows (and
therefore reduced capital expenditure), and increased economic value of properties near stormwater
schemes. For the case study system in South Australia, the yield and net present value of the scheme
would both be decreased under future climate, however, they acknowledge that urban stormwater runoff
is likely to be less affected by a drier climate than rural surface water runoff. Water saving and demand
management strategies were incorporated into a decision-support tool developed by Makropoulos et al.
(2008) and Rozos and Makropoulos (2013). This model was a demand-oriented mass balance simulation,
not incorporating hydraulic or hydrologic modelling, for the entire water cycle including wastewater
streams.
As alternative water sources are important parts of climate change adaption strategies, frameworks
developed for these sources have often been focussed on water security in future climates. Paton et al.
(2014) produced a methodology for evaluating water source alternatives under multiple future scenarios
to minimise cost and maximise water security. For nine water source alternatives with different
combinations of surface water, harvested stormwater, desalination and rainwater tanks, these objectives
were evaluated by simulating them over different future demand and climate scenarios and different
stochastic time series’ for the years 2030 and 2050. Beh et al. (2014) also investigated different water
source alternatives, however were focussed on how their implementation is sequenced. Two different
sequencing approaches were applied to the same case study and water source types used in Paton et
al. (2014); the first method was to optimise the sources used at each decision stage in sequential order,
the second method optimised the sources used in the final decision stage first, and then scheduled the
implementation of those sources. Neither of these studies considered the detailed design or operations
of the alternative water source systems. Chung and Lansey (2009) also developed a methodology for
optimal planning of WDSs, where the available sources were groundwater, surface water and recycled
wastewater. The systems were analysed over a 20 year time period, with demands increasing in line with
expected population growth and no changes to climate conditions. Chung et al. (2008) present a
mathematical model for water supply management and applied it to a hypothetical case study system to
investigate the differences between decentralized and centralized systems. Multiple sources, uses,
transportation and treatment systems can be incorporated for surface water, groundwater and recycled
wastewater sources. This does not incorporate optimization of the system, only analysis of different
systems or scenarios proposed by the user. The decision-support framework from Cai et al. (2015) can
be used for strategic planning for drought mitigation in agricultural systems under climate change. A range
of options such as infiltration ponds, parallel terraces, irrigation triggering thresholds and irrigation water
sources are available to be implemented in multiple decision stages. The performance of each possible
solution is evaluated based on three objectives; minimizing cost of drought preparedness and mitigation,
maximising agricultural production, and maximizing low flows for ecosystem conservation.
2.3 Genetic Algorithm Optimization
GAs are a robust and efficient optimization method that have been applied to many different applications,
including various water resources problems (Nicklow et al. 2010). From their first application in 1989,
Goldberg noted their desirability compared to traditional optimization techniques stemmed from four
significant differences; they work with coded representations of the solution parameters, not the
parameters themselves; they search from a population of points, not a single point; they use performance
information as the objective function, not derivatives or other system equations; and, they use probabilistic
10 |
ADE | Literature Review
rather than deterministic transition rules. Since then, they have been shown to perform very well in water
resources applications in many studies. Simpson et al. (1994) compared GAs to other optimization
techniques for pipe network design, and found that they performed better in regards to final solution
optimality and iterative efficiency. Wang et al. 2015 compared the performance several different multi-
objective evolutionary algorithms (of which GAs are a sub-set). They found GAs, in particular the non-
dominated sorting algorithm II (NSGA-II, introduced in Deb et al. 2002), performed well compared to the
other algorithms for twelve benchmark WDS design problems. Many different GAs have been developed,
and NSGA-II has been shown to perform well compared to other algorithms on multiple occasions (Barán
et al. 2005, Reed et al. 2013, Wang et al. 2015).
The basic premise of GAs is that they find (near) global optimal solutions using processes akin to natural
selection. Solutions are coded as ‘strings’ which contain decision variables. Each solution has a different
set of decision variable values. An initial population of solutions is generated at random, and the ‘fitness’
of each solution evaluated using the objective function(s). Solutions then undergo processes of selection,
crossover and mutation to produce the next generation (Figure 2.2). The fitness of each solution is
evaluated again, and the process repeated for a number of generations to converge to the optimal solution
(Goldberg 1994). The selection process randomly pairs up solutions and takes the fittest (best, for
example, minimum cost) solutions through to the next step, this is done twice, so that the number of
solutions in the population remains the same. This means that two copies of the best solution and zero
copies of the worst solution will go through to the next step. All other solutions will have either zero, one
or two copies go through, depending on their fitness values and which solutions are paired together. The
solutions that make it through the selection process are then randomly paired again for crossover. Each
pair may or may not actually undergo the crossover process, depending on the probability of crossover,
which is generally between 70 and 100%. Pairs that are selected for crossover, will then have parts of
their string swapped from a randomly selection position. The final operator is mutation, which occurs with
a much lower probability, generally less than 10%. Each gene in the string may or may not be changed
to a random value depending on this probability of mutation (Simpson et al. 1994). Constraints on the
system (such as minimum pressures for WDSs) are generally taken into account in one of two ways. The
first way is to add a penalty cost to the objective function, with the magnitude of the cost being relative to
the magnitude of the constraint violation (this could be in a linear, exponential or other type of function).
The second way is by a process called constraint tournament selection (Deb et al. 2002, Wu et al. 2010b).
When two solutions are paired up during selection, there are three possible scenarios; firstly, that both
solutions are feasible, in which case the one with higher fitness will go through; if one solution is feasible,
and one infeasible, the feasible solution will be selected regardless of their fitness values; finally if both
solutions are infeasible, the solution that violates the constraints least is selected. This type of selection
removes the need to determine an appropriate for a penalty cost value or formula.
When there are multiple objectives, the fitness evaluation and selection process is more complicated.
Multiple objectives may be combined into a single objective function using weights to normalize the values
and place different levels of importance on different objectives. Alternatively, each objective may have its
own objective function, which are then treated separately. This means that a different selection method is
required to take into account the different objectives. One such method is non-dominated sorting; when a
multi-objective algorithm is used, multiple optimal solutions are found, termed ‘Pareto’ optimal or ‘non-
dominated’ solutions. Rather than converging to a single global optimum, the algorithm converges to a
Pareto front (for two objectives, for three objectives it is a surface). Solutions on the Pareto front cannot
be improved in all objectives at the same time (Kasprzyk et al. 2012). For example, in an optimization to
minimize pumping cost and maximize the volume harvested by a water system, to decrease the cost of a
solution, the volume harvested must also decrease (the inverse of the volume harvested increases), and
11 |
ADE | Literature Review
to increase the volume harvested, the cost must also increase (Figure 2.3). Rather than comparing fitness
values of potential solutions as in a single objective algorithm, non-dominated sorting compares solutions
based on their ‘rank’, which is determined by how many other solutions they are dominated by. If two
solutions have the same rank, the ‘crowding distance’ will be compared in order to preserve variety in the
optimal front (Deb et al. 2002). NSGA-II was used in this research, and in addition to the basic GA process
shown in Figure 2.2, it implements non-dominated sorting, crowding distance comparisons and constraint
tournament selection (Figure 2.4).
$5.1M
$4.7M
$8.9M
$3.2M
$4.8M
$6.8M
$7.4M
$5.8M
Initial Evaluate Selection Next
Crossover Mutation
population fitness x2 generation
Figure 2.2: Schematic of the Genetic Algorithm process
Maximum
Feasible solution
harvesting solution
space
Infeasible solution
space
s
ts
o
c
g
n
ip
Pareto front
m
u
P Optimal solutions
Feasible solutions
Minimum cost solution
Impossible
1/ Volume harvested
solution
Figure 2.3: Example of a Pareto front
Comparisons of multi-objective and single objective optimization algorithms applied to the same problem
have been made by Savic et al. (1997) and Wu et al. (2010b). Savic et al. (1997) used GA optimization
to find optimal pump schedules to reduce energy cost and pump switches (a surrogate for maintenance
costs). Wu et al. (2010b) optimized both energy cost and greenhouse gas (GHG) emissions in WDS
design. Single-objective algorithms may be able to find some or all of the Pareto optimal solutions from a
multi-objective algorithm applied to the same problem. This can be achieved by using different weights
for the different objectives in the single objective function. The problem with this, however, is that some
information about the trade-offs between objectives is lost, and the modeller must make decisions about
the relative importance of each objective before starting the optimization. When a multi-objective algorithm
12 |
ADE | Literature Review
is used to develop a Pareto front, the trade-off information can be supplied to the decision maker, and the
relative importance of each objective examined after the optimization is performed (Savic et al. 2002).
Generate initial
population
Crossover Generate child Objective function
Mutation population evaluation
Constraint Comparison and Simulation
Handling selection model(s)
Non-dominated
Ranking
sorting
Crowding Generate global Stopping
distance population criteria met?
No
Yes
Stop
Figure 2.4: Schematic of the NSGA-II process (adapted from Wu et al. 2010b)
The primary objective for many optimization problems, in any field, is the minimization of cost, either initial,
ongoing or life-cycle. In water resources applications, other objectives such as system reliability, water
quality and environmental factors have been investigated. As climate change becomes an increasingly
serious problem for society, reduction of GHG emissions from many different industries, including the
water industry, becomes more important (Stokes et al. 2014). Stokes and Horvath (2005) undertook a
life-cycle energy analysis of two WDS case studies to determine which life-cycle stages and which water
sources (selected from imported, treated wastewater and desalination) used the most energy. Production
of electrical energy for WDSs was the biggest contributor to global warming potential throughout the life-
cycle. They also highlighted the importance of the assumed energy mix or emissions factor used in GHG
analysis. Economic costs and GHG emissions may be combined into a single objective function using a
carbon cost (for example, Marques et al. 2015), which may or may not be informed by government policy.
It is very difficult, however, to calculate the true cost of carbon emissions (Vale 2015), and as such a multi-
objective algorithm may be more appropriate. Wu et al. (2012a) investigated the sensitivity of trade-offs
between cost and GHG emissions of WDS design to the assumed electricity tariff and energy generation
mix. The assumed electricity tariff had a significant effect on the total economic costs and the optimal
solutions found, while the emissions factors affected only the GHG emissions and not the optimal
solutions on the Pareto front. If a constant GHG emissions factor is used, then the amount of GHGs
emitted is directly proportional to the electrical energy use and thus minimization of energy use can be a
surrogate for minimization of GHG emissions, such as in Ramos et al. (2011). GHG emissions factors are
variable with time, however, as the energy generation mix changes in both the short term (particularly
with renewable source reliant on weather conditions) and in the long term. An example of this is shown in
Figure 2.5 for the variation in solar photovoltaic output over one day. Generation of electricity from solar
photovoltaic panels produces less greenhouse gas emissions than traditional fossil fuel sources. As such,
when solar photovoltaic output increases during the middle of the day, overall emission factors for a region
decrease. Energy used in the middle of the day therefore results in less GHG production, and as such
energy cannot always be used as a direct surrogate for GHG emissions. Time-dependent emissions
13 |
ADE | Literature Review
factors were considered in Stokes et al. (2012b) in the optimal design of WDSs to minimize life-cycle costs
and GHG emissions. The use of time-dependent emissions factors did not affect the trade-off between
costs and GHG emissions, however, they were useful in identifying electricity usage with high emissions
intensity. The selected discount factor is another important factor that affects the trade-off between cost
and GHG emissions (Wu et al. 2010a), however, this is not applicable to studies of operations only. Stokes
et al. (2014) discussed the cost-GHG nexus for WDSs, including energy generating infrastructure, and
highlight the importance of using time-dependent emission factors and considering external factors that
influence GHG emissions such as carbon taxes and discounting.
3.5 0.605
2-
3 0.602 O
r C
e
ilp2.5 0.599 g
k
itlu M
V
P1.52 00 .. 55 99 36 ( r o tc
a
F)h W
k /q
r a 1 0.590 n oe
lo is
S s
0.5 0.587 im
E
0 0.584
0.00 4.00 8.00 12.00 16.00 20.00 24.00
Time (hr)
Figure 2.5: Daily variation in solar photovoltaic output (solid) and emission factors (dashed) (note that this figure
has been taken from Blinco et al. (2016))
For systems that utilize alternative water sources in order to reduce reliance on potable supply, the volume
of water harvested or produced by a system is also a key objective. Eusuff and Lansey (2004) considered
the amount of water reclaimed from a recycled wastewater ASR system as a sole objective. The decision
variables were the amount of recharge into the aquifer through a spreading basin (water infiltrates into
the aquifer naturally) and the rate of extraction through pumping. Various targets for water quality,
extraction well water level and residence time were analysed as constraints, with stricter targets resulting
in less water extracted. McArdle et al. (2011) performed a multi-objective optimization of a stormwater
harvesting system for potable use, considering three objectives; minimizing the present worth of capital
and operating costs (as the cost per kilolitre of water delivered to the consumer), maximizing the average
daily yield of potable water from the system, and minimizing the size of the storage in a public park to
minimize the impact on the park’s amenity. Decision variables were the capacities of the retention basin,
storage reservoir, pump, and treatment plant, and the diameter of a transfer pipe, with no operational
variables included. Without the third objective, optimal solutions would have utilized a very large reservoir
in the public park, however, to minimize the size of this reservoir, the capacity of the treatment plant can
be increased to obtain a similar yield. The cost of producing potable water from the harvested stormwater
was greater than the cost of mains water, however, this cost may increase in the future with population
growth and water security concerns. Karamouz et al. (2007) optimized an integrated surface and
groundwater system for three objectives; maximising supply for irrigation demands, minimizing pumping
costs and minimizing groundwater level fluctuations. If the groundwater level objective is ignored, water
is taken from surface sources as a priority because of the high cost of groundwater pumping. Utilizing
more groundwater, however, can help to regulate the groundwater level, which may be important in some
systems. An alternative problem formulation is minimizing the amount of potable water used, such as in
Mariano-Romaro et al. (2007) for industrial wastewater re-use.
14 |
ADE | Literature Review
2.4 Knowledge Gaps
The review of literature revealed gaps in the current knowledge that will be addressed in this thesis. With
regard to pumping operations, complex operating rules such as those utilising variable trigger levels or
combined trigger levels and scheduling have not been extensively analysed previously. The new
EPANET2 capability for optimization of rule-based controls allows these more complex control types to
be considered in optimization problems. This gap is addressed by Objective 2 (Section 1.2) and Chapter
4 (Publication 1), which optimises both simple and complex pump operating controls for two case study
systems. Another gap is the consideration of GHG emissions, which has previously been considered
using energy as a surrogate, or only with design decision variables, rather than operational decision
variables. This is covered by Objective 6 and also in Chapter 4, which specifically optimises GHG
emissions and energy use separately for pump operations.
Previous analysis and optimization of alternative water source systems has been generally focussed on
specific systems, with broad frameworks and methodologies not considered. Objective 1 covers the
development of a framework to optimize alternative water source systems and this is presented in
Chapter 5 (Publication 2). Application of this framework to two case studies – a harvested stormwater
system and an integrated alternative water source supply system – is also included in Chapter 5.
Optimization of detailed pump operations and consideration of hydraulics has often been left out of studies
on alternative water sources. This gap is addressed in Objective 3 and Chapter 6 (Publication 3), which
focuses on the harvested stormwater case study and utilises EPANET2 for detailed pump energy use and
hydraulic calculations.
Many studies also perform only optimization, without in depth simulation or sensitivity analysis performed
prior to carrying out the optimization study. Pre-optimization analysis by extensive simulation analysis can
provide vital information for the formulation of the optimization problem. The size of the optimization
problem can be reduced by identifying infeasible or undesirable options by simulation of the system.
Sensitivity analysis can also be combined with optimisation in order to assess the robustness of the
system to different conditions. This gap is addressed by Objectives 4 and 5, as well as in Chapter 5
which performs a simulation analysis of a harvested stormwater system and in Chapter 6 which then
covers sensitivity analysis and optimization of the same system.
15 |
ADE | Chapter 3 Synopsis of Publications
This chapter discusses the contributions made by the three publications presented in this thesis, their
connections, and how they address the objectives of the work. The overall aim of this research is to
develop and apply methodologies for optimizing complex pumping operations to systems that use
alternative water sources. EPANET2 hydraulic simulation software is utilised in all of the publications,
and this guarantees the conservation of energy and mass, which are constraints of the pump operations
optimization problem. Figure 3.1 shows the contributions of the publications to the six specific objectives
listed in Section 1.2. Publication 1 investigated five different types of pumping regimes using EPANET2
rule-based controls (Objective 2). These regimes were optimized and compared for two potable case
study networks, considering objectives of minimizing pump energy costs and minimizing GHG
emissions from pumping (Objective 6). Publication 2 presents a framework for the optimization of water
supply and distribution systems that use alternative water sources (Objective 1). Sensitivity analysis of
variables that have some uncertainty is also discussed (Objective 5) and two case studies demonstrate
the application of the framework. Finally, Publication 3 applies the framework methodology from
Publication 2 and the use of rule-based controls from Publication 1 to a harvested stormwater system
(Objective 3). It includes extensive analysis of the case study system (Objective 4) and sensitivity
analysis of the operation of the system to pump and tank sizing (Objective 5).
Publication 1 Publication 2
Complex pump operating rules in potable Framework for optimization of alternative water
WDSs source systems
2.* EPANET2 6. Cost and GHG 5. Sensitivity
1. Framework
rule-based controls trade-off analysis
Publication 3
Pump operations in a harvested stormwater system
3. Optimize harvested 4. Pre-optimization
5. Sensitivity analysis
stormwater operations analysis
*Numbers refer to objectives listed in Section 1.2
Figure 3.1: Connection between publications and their contributions to the research objectives
Optimization techniques have been extensively applied to pump operations problem for WDSs, both
using trigger levels and scheduling. Previously, the ability to optimize complex operating rules using
hydraulic simulation software was limited; simple trigger levels or scheduled could be controlled,
however, trigger levels that vary with time could not. New developments for EPANET software to
optimize the more complex rule-based controls were presented and tested in Marchi et al. (2016b). The
main objectives considered in many optimization studies has been cost of energy use, system efficiency
and reliability. Often, design and operation of a system have been optimized together, and in some of
these cases, GHG emissions have been considered as an objective. GHG emissions are becoming a
more important objective, as many water system managers have sustainability goals to consider. For
existing systems, the majority of GHG emissions come from electrical energy use for pumping
operations. Many previous studies focussing on GHG emissions have considered design decision
variables rather than operational changes. Reducing the GHG emissions of existing systems through
operational decision variables has not been extensively researched.
17 |
ADE | Synopsis of Publications
Publication 1 compares different operational pumping strategies, using both simple controls and
complex controls, for cost and GHG emissions of pumping operations in potable WDSs. The new
EPANET programmer’s toolkit to alter rule-based controls was applied to consider five different types of
pump operating regimes; (1) lower and upper trigger levels; (2) a reduced upper trigger level; (3)
combined trigger levels and scheduling; (4) variable trigger levels; and (5) variable speed pump
scheduling (Objective 2). A single-objective genetic algorithm as used to optimize the cost and GHG
emissions from pumping separately (Objective 5). Costs were calculated based on the energy use of the
pumps across a 24-hour period with a peak and off-peak electricity tariff. Energy use of the pumps was
converted to GHG emissions based on emissions factors of energy generation technology (in kg of CO
2
equivalent per kWh). The emissions factors were based on the current South Australia energy
generation breakdown, with some variation over the 24-hour simulation period based on the varying
contribution of solar photovoltaic energy over a day. Two case study WDSs were used to compare the
performance of the different pump operating regimes; a hypothetical one-pipe network, and a portion of
the real-life South Australian WDS. Time-based scheduling operating strategies were found to perform
better than the other regimes for both case studies. Significant cost savings were achieved for the South
Australian system compared to its current operation.
Applying the methodologies that have been developed for and used on potable WDSs to alternative
water source systems requires additional complexities to be taken into account. Traditional natural
catchment supplies have often been split between hydrological analysis of the supply side, and
hydraulic analysis of the demand side, with large storages delineating the two. Analysis and
optimization of WDSs has assumed that there is always enough water available in the supply reservoir
or there is a set discharge available from a water treatment plant. For alternative water source systems,
this is not always the case, and it is important to analyse the supply from the catchment for sources
such as stormwater and groundwater to know when the alternative water can be supplied, and when
potable back-up should be used. Alternative water source systems also use infrastructure and
technology that are not often part of a potable WDS and need to be modelled. This includes
components such as wetlands bioretention basins in stormwater systems, bores in groundwater
systems and small-scale treatment technologies in decentralized systems. Previous methodologies and
frameworks for traditional potable WDSs therefore do not have the modelling capability required by
alternative water source systems. Those developed for alternative water source systems, however, are
often not generalized to many different water source types, and do not include detailed consideration of
pumping and hydraulics.
Publication 2 presents a framework for the optimization of water supply and distribution systems that
use alternative water sources along with a detailed discussion of the components and key variables
(Objective 1). The options component describes the potential decision variables, both design and
operational; the infrastructure component describes the physical components of the system to be
modelled, including energy infrastructure that affects the evaluation of electricity costs and emissions;
the analysis component describes the simulation of each potential system configuration and how it is
evaluated against objectives and constraints; there is also a government policy component that covers
policies from regulating bodies that may affect other parts of the framework. These all exist within an
optimization algorithm structure, which would analyse and evaluate different potential solutions to find
those that meet the constraints and perform best in terms of the objectives. Sensitivity analysis of
demand, rainfall and streamflow, electricity and GHG emissions, discount rates, and climate change is
also discussed (Objective 5). Two case study systems are used to illustrate how the framework can be
applied to minimize the cost of water system operations. The first – the Ridge Park Managed Aquifer
Recharge System – is a harvested stormwater and managed aquifer recharge (MAR) that supplies non-
18 |
ADE | Synopsis of Publications
potable water for irrigation of public reserves. This system can be split into seasonal operations; winter
stormwater harvesting and injection, and summer extraction and irrigation. The current operation of this
system is analysed by hydraulic simulation in order to formulate an optimization of pumping operations.
The second case study – the Orange Integrated Supply System – utilizes several different water
sources; natural catchment, harvested stormwater, groundwater and imported water (from an adjacent
catchment) to supply potable water to over 35, 000 people. In this system, it is important not to waste
water by pumping from one of the three alternative sources only to have rain fill the natural catchment
reservoirs, and this is considered by including an objective to minimize spills. Optimization of pumping
operations for this case study focusses on reducing pump energy use. Figure 3.2 and Figure 3.3
demonstrate how these case studies fit in to the developed framework. The elements highlighted in the
framework diagrams are those that are considered by each case study. Note that while optimization of
the Ridge Park Case Study is not performed in Publication 2, it is covered in Publication 3 and therefore
is highlighted in Figure 3.2.
As for potable WDSs, pumping energy is a large contributor to costs in alternative water sources
systems, including harvested stormwater schemes. The focus of optimization of stormwater systems
has been on their design, rather than operation. Harvested stormwater schemes often include multiple
pumps between multiple storages, which can result in complex operating rules. The status of each
pump relies on the level in more than one storage, and the level in each storage relies on the status of
more than one pump. Optimization of complex pump operating rules, as in Publication 1, can be applied
to harvested stormwater systems, however, additional modelling capability needs to be incorporated
and different constraints and objectives considered, as discussed in Publication 2. Expanding current
methods for optimizing pump operations in potable WDS to alternative water source systems will allow
these systems to perform better and become more a desirable option to water system managers. As
climate change and population growth raise water security concerns into the future, alternative water
sources will become more necessary, and as such reducing their cost of operation is important.
Publication 3 explores the operation of a harvested stormwater case study system from South Australia
both through simulation sensitivity analysis (Objective 5) and multi-objective optimization. The system
has distinct winter and summer operational seasons; harvesting water from an urban creek, treating and
injecting it into an aquifer during winter, and extracting water from the aquifer for irrigation of public
reserves during summer (Objective 3). Most of the irrigation sites are on a gravity fed line, with the three
closest to the harvest site, and highest in elevation, are on a pressure line. There are four pumps in the
system, two used only in winter, one used in both winter and summer, and one used only in summer
(Objective 3). Significant analysis of the system was preformed prior to optimization, to determine the
current operation with different possible inflows, and determine the most appropriate way to model some
of the components (Objective 4). For the winter operation, storage trigger levels were implemented as
rule-based controls in EPANET and optimized to minimise the cost of pumping and maximise the
volume of water harvested. During summer, irrigation scheduling, and the trigger levels for the bore
extraction pump were optimized to minimize pumping costs. Restrictions on the aquifer injection rate
and pressure are considered, as well as pressure and demand requirements at the various parks and
reserves. The installation of new pumps and a larger tank are considered in both the simulation
sensitivity analysis and optimization. Recommendations from the results of the optimization were to
install new pumps with lower flow rates and better efficiencies, to utilize the full height of the storages by
using wider trigger levels and to irrigate all reserves on the pressure line together.
19 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
Abstract
A single-objective optimization model has been developed for water distribution system (WDS) pumping
operations, considering five different types of pump operating regimes. These regimes use tank trigger
levels, scheduling, and a combination of both to control pumps. A new toolkit development to alter rule-
based controls in hydraulic simulation software has allowed more complex pump operating regimes than
have previously been considered to be optimized. The performance of each of the regimes is compared
with respect to two different objectives: cost and greenhouse gas (GHG) emissions, which were optimized
separately to allow the comparison of regimes to be made more clearly. Two case study networks,
including one that represents a segment of the South Australian WDS, illustrate the effectiveness of the
model. Time-based scheduling operating strategies were found to perform better than the other types of
pump operating regimes. Significant cost savings were achieved for the South Australian case study
network compared with its current operation.
26 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
4.1 Introduction
Energy costs can account for up to 65% of a water utility’s operating budget (Boulos et al. 2001), and as
such optimizing the cost of energy used for pumping will have significant benefits. Previous investigations
of optimal pump operating strategies have generally been restricted to either lower and upper tank trigger
levels or scheduling. Consideration of more complex pump operating regimes, for example, using trigger
levels that vary throughout the day or combining trigger levels and scheduling, has been restricted in part
by simulation model capabilities. A modification of the existing EPANET2 toolkit (Rossman 2000) has
been developed by Marchi et al. (2016b) in order to modify rule-base controls. This new toolkit is called
“EPANET2-ETTAR” (EPANET2 Toolkit to Alter Rules) and allows more complex pump operating regimes
to be optimized. Human-induced climate change presents a serious global risk and action to mitigate the
impact by reducing greenhouse gas (GHG) emissions is important. Production of electrical energy for
water distribution system (WDS) pumping operations is the biggest contributor to GHG emissions from
the water industry (Stokes and Horvath 2006; Wu et al. 2013).
This paper describes the development of a single-objective genetic algorithm (GA) optimization model for
WDS pump operations integrating EPANET2 (including EPANET2-ETTAR) and a Microsoft Excel
interface. The performance of five different types of pump operating regimes, including trigger levels that
vary throughout the day and combined trigger levels and scheduling, is compared with respect to either
the minimization of cost or the minimization of GHG emissions. The model is applied to two different case
studies, a hypothetical one-pipe network and a real-life network from South Australia. In the second case
study, two different pump sizes are considered and the results compared.
4.2 Literature Review
Efficient operation of WDSs can be achieved in several ways. The first step is to optimize the design of
pumps and infrastructure, then, for existing or designed systems, pump operating rules can be optimized.
Other strategies include recovering energy that would otherwise be dissipated using mini-hydro systems
(Carravetta et al. 2013b; Fecarotta et al. 2015), reducing leakage to reduce pump and water treatment
energy requirements (Giustolisi et al. 2013) and pump maintenance or replacements. There are many
different objectives that can be considered to achieve efficient WDS operation, with the most common
being to minimize the cost of electrical energy use. GHG emissions, based on energy use, or simply
energy use itself can be used as environmental impact objectives (Simpson 2009). Water quality can be
addressed by minimizing water age, which can be obtained from EPANET2 (Stokes et al. 2012a); pump
maintenance cost, represented by pump switches, could be formulated as an objective (López-Ibáñez et
al. 2005) or as a constraint (Lansey and Awumah 1994); system effectiveness (Carravetta et al. 2013a),
resilience (Prasad and Park 2003), and leak reduction (Giustolisi et al. 2015) can also be used as
objectives to improve the performance of WDSs.
The research presented in this paper focuses on the optimization of pump operating rules and the
comparison of different types of pump operating structures. The case studies investigated are existing
systems, and therefore no design optimization is considered. Objectives of pumping electricity cost and
GHG emissions are considered separately and the characteristics of the optimal operating strategies for
the objectives are compared. Multiobjective optimization of cost and GHG emissions for WDSs has been
extensively covered in Wu et al. (2010a, b, 2011, 2012a, b, 2013) and Stokes et al. (2012b, c, 2014). This
research is different in that it considers the effect of the different pump operating regimes on each
objective individually. WDSs are often required to perform under different conditions, including different
demands (e.g., seasonal and daily variations), emergencies (such as fires), and failure scenarios (such
as power outages or pipe breaks), all of which have some uncertainty associated with them. Goryashko
27 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
and Nemisrovski (2014) use stochastic methods to find optimal operating strategies that are robust to
different demand scenarios, while Basupi and Kapelan (2015) combine Monte Carlo analysis with GA
optimization for the WDS design problem. Analysis of emergency conditions and system failure in
optimization has been much more widely applied to the design problem (e.g., Morley et al. 2012) while,
for pumping operations, the use of a constraint on the minimum tank level or an emergency reserve
storage is usually used to guarantee a reliable service.
Optimization of pump operations is highly complex due to a large number of possible pump operating
strategies, variable electricity price, and fluctuating consumer demands. Operational policies are also
subject to several constraints, including acceptable levels of water in storage tanks, maximum pumped
volumes, long-term tank level balancing, nodal pressure limits, and maximum pipe velocities. Previous
studies have usually been restricted to using either trigger levels (Paschke et al. 2001; Stokes et al. 2012b)
or scheduling (Mackle et al. 1995; Goryashko and Nemisrovski 2014) and have not considered more
complex operations such as trigger levels that vary throughout the day or combinations of trigger levels
and scheduling. Lower and upper trigger levels represent the tank levels at which the pump(s) will turn on
or off, respectively (when pumping to a downstream tank). Pump scheduling involves a set of temporal
rules indicating when pumps should be switched on or off during the day. Scheduling requires an accurate
estimation or a forecast of the expected daily water demand. Kazantzis et al. (2002) combined the use of
trigger levels and scheduling, however, the trigger levels were fixed, and only the scheduling variables
optimized. In EPANET2 (Rossman 2000), only simple controls (used for trigger levels) and pump patterns
(used for scheduling) can be altered through the programmer’s toolkit (which can be used to trial different
potential solutions within, say, a genetic algorithm optimization framework), and rule-based controls that
are required for more complex operating regimes cannot be changed via the current toolkit. EPANET2-
ETTAR gives access to these rule-based controls, therefore allowing more complex pump operating
regimes to be considered in the pumping optimization process.
When a peak and off-peak electricity tariff structure applies, operational costs will be minimized by
reducing the amount of pumping in the peak electricity period and deferring this pumping to the off-peak
period. Operational costs will also be reduced by reducing the static head and by increasing the efficiency
of the operating point. Maximizing the amount of off-peak electricity pumping can generally be achieved
when the tank water level is at its maximum at the beginning of the peak period and at its lowest allowable
level at the end of the peak period (Mackle et al. 1995; Kazantzis et al. 2002). A future approach, primarily
concerned with GHG emissions, may be to pump steadily throughout the day with a variable speed pump
(VSP), or in response to demands rather than electricity prices, with reduced energy through the use of
slower velocities leading to a smaller friction head loss (Simpson 2009).
To properly account for the GHG emissions of WDSs, the sources of electricity should be identified
because each will have different GHG emissions per unit of energy produced (Dandy et al. 2006). An
emission factor is used to convert energy use to GHG emissions, considering all types of GHGs and their
global warming potential as an equivalent mass of CO (CO -eq). Previous studies have used an average
2 2
GHG emission factor value for the region, including Dandy et al. (2006) and Wu et al. (2010a, b). Stokes
et al. (2012b) took into account time-varying emission factors in their optimization of water distribution
system design and operation. This identified high emission intensity electricity use and helped to reduce
operational GHG emissions. The objectives of cost and GHG emissions may be aligned if no variation in
electricity tariffs or emission factors is considered. When variations in these factors are taken into account,
times with lower electricity prices will not necessarily coincide with times of lower emission factors, so
optimal solutions for the two objectives will be different.
28 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
GAs represent an efficient method for the optimization of nonlinear problems, particularly when applied to
complex WDSs. These algorithms are a population-based optimization technique that use coded
representations of solutions (Goldberg 1989). After generating a random initial population, the GA
determines the fitness of each potential solution by simulating them and evaluating an objective function.
In many optimization problems, the objective function is based on cost, but it can also be formulated for
other objectives. All solutions then go through GA operators based on evolutionary principles—typically
selection, crossover, and mutation–to produce the next generation of solutions (Goldberg 1994). This
process is repeated to converge on optimal or near-optimal solutions. When applied to the optimization
of WDSs, GAs have been found to perform significantly better than other optimization techniques in areas
of final solution optimality and iterative efficiency and are still competitive with other optimization methods
today (Simpson et al. 1994; Wang et al. 2015).
4.3 Methodology
4.3.1 Optimization Model Formulation
The aim of this research was to compare the performance of five different pump operating control cases
and the characteristics of their optimal solutions. To achieve this aim, a single-objective optimization
model was developed, linking a GA with EPANET2- ETTAR and a Microsoft Excel Interface. EPANET2-
ETTAR was used to simulate the different potential solutions from the GA in order to provide information
about their performance relative to the objective function and constraints. The interface allowed the
optimization parameters, decision variables, choice tables, and other inputs to be changed and
customized for different networks. A single-objective GA with tournament selection, a choice of one- or
two-point crossover, and bitwise mutation was used. Trigger level cases, with a small number of decision
variables, used one-point crossover with a crossover probability of 0.8, a mutation probability of 0.05, 200
generations, and a population size of 200. Scheduling cases, with a large number of decision variables,
used two-point crossover with a crossover probability of 0.7, a mutation probability of 0.02, 400
generations, and a population size of 300.
Wherever possible, full enumeration of the search space was used in preference to the genetic algorithm
optimization. Two different objective functions were considered separately: cost and GHG emissions. The
value of each objective function was calculated in terms of units per volume of water pumped to remove
any bias between solutions that pumped different volumes of water over the day. For the cost optimization,
the objective function was dependent on the energy use, electricity tariff rates, and the volume of water
pumped over the whole day as given by Eq. (4.1)
∑ 𝑇 ×𝐸
𝑂𝐶 = 𝑖 𝑖 𝑖 (4.1)
𝑉
where OC = operational cost (dollars/m3); T = electricity tariff for each time step i (dollars/kWh); E =
i i
energy consumption for each time step i (kWh); and V = total volume pumped (m3) during the time
simulation period. EPANET2-ETTAR was utilized to determine energy use for each time period as well
as the volume of water pumped. In this research, a two-part electricity tariff has been considered, however,
the pattern for the electricity tariff could easily be altered to consider other, perhaps more complex, tariff
structures, such as a multipart tariff (more than two periods). In addition, a monthly peak energy demand
charge (that is, an additional charge for the maximum kilowatt usage) could also be included if desired.
An electricity price pattern can be specified in EPANET2, as well as a demand charge variable, which
may apply if there is a monthly peak energy demand charge. Electricity costs were based on a
representative South Australian tariff; a peak electricity price of 22 c/kWh (c = cents) between 7 a.m. and
11 p.m., and an off-peak electricity price of 9 c=kWh from 11 p.m. to 7 a.m.
29 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
The objective function for GHG emissions was based on the distribution of emission factors throughout
the day and the energy used in each time period as given by Eq. (4.2)
∑ 𝐹×𝐸
𝑂𝐺𝐻𝐺 = 𝑖 𝑖 𝑖 (4.2)
𝑉
where OGHG = operational GHG emissions (kgCO -eq/m3); F = emissions factor at each time step i
2 i
(kgCO -eq/kWh); and E = energy at each time step i (kWh), which ranged from 0 to 23 for hourly time
2 i
increments. Emission factor data were collated from Dey and Lenzen (2000), Lenzen (2008), and Evans
et al. (2010) in order to take into account the varying contributions to GHG emissions from different energy
technologies. To calculate the overall emission factor, South Australia’s current energy sources, mainly
gas, brown coal, and wind (Australian Energy Market Operator 2011), have been used. The emission
factors were also adjusted to account for the variation in output from solar photovoltaic systems
throughout the day and this output was greatest during the middle of the day (Figure 4.1). The contribution
of each energy source at every hour was adjusted depending on the solar photovoltaic multipliers to give
a daily variation in emission factors, which were lowest in the middle of the day (Figure 4.1). Minimization
of energy consumption was also available in the model and acted as a surrogate for optimization of cost
or GHG emissions where no daily variation in electricity tariffs or emissions factors was present.
3.5 0.605
2-
3 0.602 O
r C
e
ilp2.5 0.599 g
k
itlu M
V
P1.52 00 .. 55 99 36 ( r o tc
a
F)h W
k /q
r a 1 0.590 n oe
lo is
S s
0.5 0.587 im
E
0 0.584
0.00 4.00 8.00 12.00 16.00 20.00 24.00
Time (hr)
Figure 4.1: Daily variation in solar photovoltaic output (solid) and emission factors (dashed)
A number of constraints could be used in the optimization process, with penalties added to the objective
function in the case of constraint violation. In addition to pressure, velocity, and head loss constraints, a
minimum tank level may be specified to account for emergency and dead storages. There was also a tank
balancing constraint, formulated as the maximum allowable difference between the storage tank’s start
and end level each day, and this could be used to prevent depletion of the water in the tank at the end of
the simulation period. The maximum number of pump switches to occur within a 24-h period may also be
specified, which could be used to address issues of pump maintenance costs.
4.3.2 Pump Operating Control Cases
Optimization of five distinct pump operating control cases was considered: (1) Case A, lower and upper
trigger levels; (2) Case B, a reduced upper trigger level; (3) Case C, combined trigger levels and
scheduling; (4) Case D, variable trigger levels; and (5) Case E, variable speed pump scheduling. The
pump operation was optimized over a period of 24 h, with the simulation period beginning at the start of
the off-peak tariff period and the water level in the tank being at its lowest allowable level. This serves as
a known starting point for an optimal solution and also means that the final water level of the tank is likely
to be close to the initial level as less pumping will benefit either of the objective functions. The available
decision variables and constraints for each pumping control case are summarized in Table 4.1.
30 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
Table 4.1: Summary of decision variables and constraints for each control case
Case Decision Variables Constraints
A Lower trigger level; upper trigger level
Minimum tank level
B Lower trigger level; reduced upper trigger level; upper trigger level
Tank balancing tolerance
Lower trigger level; upper trigger level; scheduled pump start(s); scheduled pump
C Maximum pump switches
stop(s)
Max./min. nodal pressures
Peak lower trigger level; peak upper trigger level; off-peak lower trigger level; off-
D Max./min. pipe velocities
peak upper trigger level
Max./min. pipe headloss
E Pump speed multiplier(s) (number depends on time interval)
Control Case A optimized two decision variables—the lower and upper trigger levels in a downstream
tank that determined when a pump would be switched on and off, respectively. While trigger levels are
effective at keeping the water level in a tank within a certain operating range, there are both advantages
and disadvantages to different trigger level operating strategies. Increasing either trigger level will
increase the average static head of the system and therefore requires the pump to expend more energy
to pump the same volume of water to the tank. A lower value of the upper trigger level may increase the
amount of pumping required in the peak electricity tariff period because the tank will not be full at the start
of this period, and hence may increase costs. The closer the trigger levels are to each other, the more
times the pump will switch on and off during the day, which will increase general wear and tear of the
pumps. Additionally, having both trigger levels or just the lower trigger level closer to the minimum
allowable tank level may jeopardize the system’s capability to meet demand requirements. In times of
extremely high demand, the rate at which the tank is draining may exceed the maximum pumping
capacity, resulting in overall depletion of the tank volume even with the pump switched on. In these
circumstances, if the trigger levels are too low, the water level in the tank may fall below the minimum
allowable level.
A reduced upper trigger level was considered in Control Case B, which implemented EPANET2-ETTAR
for optimization of rule-based controls. This model had three decision variables: a lower trigger level, an
upper trigger level, and a reduced upper trigger level. During most of the 24-h simulation period, a reduced
upper trigger level was permitted in order to reduce the static head of the system. There was a user-
selected switch time before the start of the peak period at which the control would swap to the ultimate
upper trigger level in order to fill the tank before the peak period.
Control Case C combined the use of tank trigger levels and pump scheduling. There were two trigger
level decision variables—an upper and lower trigger level—which governed most of the pump operation.
In addition to this, multiple time-based scheduling decision variables were also included that would specify
a time for pump starts and pump stops. These time-based decision variables allow the tank water level
criteria at the end of each tariff period [as identified by Mackle et al. (1995) and Kazantzis et al. (2002)] to
be met where trigger levels alone cannot achieve this. For example, if the trigger levels in a particular
network were such that the tank was draining at the end of the off-peak period, a scheduled pump start
was added so that the tank is full at the start of the peak period. If the tank is filling at the end of the peak
period, a scheduled pump stop was added to ensure the tank would be at its lowest allowable level at the
end of the peak period and therefore avoid excess peak pumping.
Control Case D allowed for different trigger level sets for the peak and off-peak periods and this also
utilized the EPANET2- ETTAR toolkit. There were four decision variables—an upper and lower trigger
level in the peak period and an upper and lower trigger level in the off-peak period. In order to reduce the
pumping cost, the two trigger levels used for the off-peak period will be higher than the two trigger levels
used for the peak period because this allows the tank level to be closer to full at the beginning of the peak
31 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
tariff period and close to the minimum allowable tank level at the beginning of the off-peak period. As
suggested by Kazantzis et al. (2002), in order to optimize costs the tank should be at its minimum level
at the end of the peak period and at its maximum level at the start of the peak period. The two different
sets of trigger levels also allow for the reduction of the static head (and therefore energy use) during the
period of higher electricity cost.
VSPs were incorporated into Control Case E, which optimized pump scheduling regimes. The decision
variables in this model were the pump speed multipliers at each time interval. If fixed speed pumps (FSPs)
were used, the only possible values for the pump speed multipliers would be 0 or 1. For VSPs, additional
choices for the multipliers could range from 0.85–1.0 (as well as 0 for when the pump is off). The minimum
pump speed multipliers calculated for the specific case studies take into account the guidelines by Marchi
et al. (2012): (1) the minimum relative speed of the pump is larger than 0.7 so that the affinity laws can
be used to predict the pump efficiency curve with reasonable accuracy, and (2) the shutoff head of the
pump curve at the reduced speed is still higher than the static head of the system in order to deliver a
flow larger than zero. In particular, the lower limit (0.85 in this case) depends on the pump shutoff head
relative to the maximum system static head. Variable speed drive efficiency is not taken into account and
this could affect the energy use of VSP solutions (Walski et al. 2003). When choosing a VSP for a
particular system, the overall efficiency, including the variable speed drive efficiency and motor efficiency,
should be taken into account. The time interval for the simulation of the pump schedule could be modified
to reflect different demand patterns and pumping restrictions or requirements. For example, half-hourly
time intervals would result in 48 decision variables, which could increase operational flexibility but also
could increase optimization run times and effectiveness compared with hourly time intervals with only 24
decision variables. For systems with multiple pumps, a larger time interval may need to be used because
otherwise the number of decision variables may easily become excessive, leading to long optimization
run times and a larger search space, making finding the optimal solution more difficult.
4.4 Results
4.4.1 Case Study 1: One-Pipe Network
The models were initially used to analyze a one-pipe network introduced by Wu et al. (2010a), who
performed a multiobjective optimization for the pump size and pipe diameter of the network, finding eight
nondominated solutions in terms of capital and operating costs and GHG emissions. A design solution
that represented an acceptable trade-off between costs and GHG emissions was used in this research
(Figure 4.2 shows the network configuration). The network pumped water from an upstream reservoir to
a downstream tank, which supplied an average peak day demand of 80 L/s. A diameter of 20 m was
assumed for the downstream circular tank. Potential trigger level values for this network ranged from 1.0
to 5.0 m, with an increment of 0.2 m. The minimum possible trigger level value accounted for dead storage
and emergency reserves. VSP multipliers considered were between 0.85 and 1.0 in 0.05 increments
(Table 4.2). The minimum feasible VSP multiplier was determined using the first pump affinity law
relationship between pump head (H ) and speed (N) [Eq. (4.3)]. Pump speed can be reduced to a point
P
where the shutoff head of the pump is equal to the static head of the system. At full speed [1,475
revolutions per minute (rpm)], the pump shutoff head is 143 m (H ) and the static head of the system
P1
when the tank is full is 100 m (H ). Applying Eq. (4.4) gives a minimum pump speed multiplier (N ) of
P2 2
0.84; to be conservative, a minimum value of 0.85 is considered (equivalent to approximately 1,254 rpm)
𝐻𝑃1
=
(𝑁1)2
(4.3)
𝐻𝑃2 𝑁2
32 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
if N
1
= 1 (full speed) then 𝑁
2
=
√𝐻𝑃2
(4.4)
𝐻𝑃1
EL 95.0 m
L = 1500 m
D = 375 mm
EL 0.0 m ε = 0.25 mm
Figure 4.2: One-pipe network
Control Case A
Cost Minimization. When optimizing pump operating Control Case A, a lower trigger level of 1.0 m and
an upper trigger level of 5.0 m was the best solution in terms of cost (Table 4.3). Because there were only
two decision variables, each with 21 possible values (using increments of 0.2 m), the total number of
possible solutions was 212 = 441. Complete enumeration of the problem was performed and confirmed
this result. The second-best through to the sixth-best solutions as presented in Table 4.3 show the same
characteristic of having the trigger levels far apart, allowing maximum off-peak pumping. Solutions 7, 8,
and 10 reduce energy use and therefore cost by reducing the static head of the system. These solutions
all had a trigger level range of 1.6 m, with different lower and upper trigger levels. This trigger level range
allowed the tank to half-fill twice during the off-peak period while also maintaining a lower water level than
the first six solutions (Figure 4.3). As can be seen in the “Energy” column, the seventh solution had the
lowest energy use per volume of water pumped from the cost optimization solutions. It had a greater cost
per volume pumped because there is a greater percentage of energy being used in the peak period
compared with the first six solutions (“Peak energy” and “Off-peak energy” columns). This indicates that
for this network, the effect of the peak and off-peak tariff prices on the cost is greater than the effect of
reducing the static head.
Table 4.2: Summary of choices and constraints applied to each case study
Decision Variable / Constraint One-Pipe Network South Australian Network
Trigger levels (m) (Cases A-D) 1.0-5.0 m, 0.2 m increment 4.0-7.9 m, 0.1 m increment
First pump start (Case C) 3am-7am, 5 min. increment 3am-7am, 5 min. increment
Second pump start (Case C) 4pm-10pm, 5 min. increment -
Pump stop (Case C) 10pm-11:30pm, 5 min. increment 6pm-10pm, 5 min. increment
Pump speed multipliers (Case E) 0.85-1.0, 0.05 increment 0.88-1.0, 0.04 increment
Minimum tank level (m) None, 0.8 m, 1.0 m 2.5 m, 4.0 m
Tank balancing tolerance (m) None, 0.5 m None, 0.1 m, 0.5 m
Maximum pump switches 12, 96 12, 96
Min./max. nodal pressures (m) - None, 20/120 m
Min./max. pipe velocities (m/s) - None, 0/5 m/s
Min./max. pipe headloss (m/km) - None, 0/50 m/km
The solutions represented in Table 4.3 and Figure 4.3 did not have a minimum tank level constraint
enforced, which allowed the water level to fall significantly below the lower trigger level of 1 m due to high
demands in the evening (“Minimum water level” column Table 4.3). If a minimum tank level constraint of
1 m is used, the optimal trigger levels are found to be 1.6 and 3.2 m (the 10th-best solution in Table 4.3),
which has a minimum tank level of 1.32 m, well above the constraint. If the minimum level constraint is
relaxed slightly, the optimal trigger levels are found to be 1.2 and 2.8 m (the eighth-best solution in Table
33 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
4.3). This results in a minimum tank level of 0.96 m, which may be acceptable to the decision maker. This
shows the impact of the minimum tank level in finding the optimal trigger levels.
Table 4.3: Top solutions from pump operating Control Case A optimization for the one-pipe network
Lower Upper Trigger Min.
Peak Off-peak GHGs
Cost Trigger Trigger Level Energy Water
Solution Energy Energy (kg CO -
($/m3) Level Level Range (kWh/m3) Level 2
(%) (%) eq/m3)
(m) (m) (m) (m)a
Cost: 1st 0.0683 1.0 5.0 4.0 0.3725 72.0 28.0 0.36 0.2222
Cost: 2nd 0.0688 1.0 4.8 3.8 0.3721 73.1 26.9 0.40 0.2220
Cost: 3rd 0.0690 1.2 5.0 3.8 0.3728 73.1 26.9 0.59 0.2224
Cost: 4th 0.0695 1.0 4.6 3.6 0.3718 74.5 25.5 0.48 0.2219
Cost: 5th 0.0696 1.2 4.8 3.6 0.3725 74.4 25.6 0.66 0.2223
Cost: 6th 0.0697 1.4 5.0 3.6 0.3731 74.4 25.6 0.85 0.2227
Cost: 7th 0.0698 1.0 2.6 1.6 0.3702 75.9 24.1 0.77 0.2213
Cost: 8th 0.0699 1.2 2.8 1.6 0.3708 75.8 24.2 0.96 0.2218
Cost: 9th 0.0701 1.0 4.4 3.4 0.3716 75.9 24.1 0.60 0.2218
Cost: 10th 0.0701 1.6 3.2 1.6 0.3721 75.7 24.3 1.32 0.2225
GHG: 1st 0.0721 1.0 1.2 0.2 0.3685 81.2 18.8 0.45 0.2204
aMaximum water level for each solution is equal to the upper trigger level.
Off-Peak Peak Off-Peak Peak
5.0 5.0
)m Upper Trigger Level: 5.0 m )m
( le
v e L
re34 .. 00 ( le
v e L
r34. .0
0 Upper Trigger Level: 2.6 m
ta
W2.0
e
ta2.0
k
W
n
a k
T1.0 n1.0
Lower Trigger Level: 1.0 m a Lower Trigger Level: 1.0 m
T
0.0 0.0
(a) Time (hr) Cost: 0.0683 $/m3 (b) Time (hr) Cost: 0.0698 $/m3
Figure 4.3: Daily tank level variation of the one-pipe network: cost optimization solutions: (a) pump operating
Control Case A, first solution; (b) Control Case A, seventh solution
GHG Minimization. The optimal solution for GHG emissions was different than the optimal cost solution.
The lower and upper trigger levels were as low and as close together as possible, at 1.0 and 1.2 m,
respectively (while in the cost optimal solution they were as far apart as possible), reducing the static
head. No effect due to the daily variation in GHG emission factors was observed in the optimal GHG
solution. Because the trigger levels are very close together, the pump turns on and off quite often (62
pump switches) throughout the day, with the exception of two blocks in the peak period where the pump
is on, resulting in higher costs. The seventh cost solution had lower GHG emissions than the other top 10
cost solutions (“GHGs” column of Table 4.3). Because it reduced energy use and costs by reducing the
static head as well as reducing peak pumping, it was an acceptable compromise between the cost and
GHG objectives.
Control Case B: Cost Minimization
With the addition of a reduced upper trigger level in Control Case B, the minimum operating cost was
lowered to $0.0652/m3, compared with the $0.0683/m3 for the Control Case A solution. A switch time of
2 a.m. gave the lowest cost and was able to fill the tank just before the start of the peak period at 7 a.m.
[Figure 4.4(a)]. Using a reduced upper trigger level did not benefit GHG emissions because there was no
34 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
need to fill the tank before the start of the peak period and a reduced static head could be achieved using
a low value for the upper trigger level.
Off-Peak Peak Off-Peak Peak
5.0 5.0
)m Upper Trigger Level: 5 m )m Upper Trigger Level: 5 m
( le4.0 ( le4.0
Pump Stop:
v v
e L e L Pump Start: 10.20 pm
re3.0 re3.0
5.35 am
ta W2.0 Reduced Upper Trigger Level: 2 m ta W2.0 Pump Start:
k k 6.20 pm
n n
a a
T1.0 T1.0
Lower Trigger Level: 1 m Lower Trigger Level: 1.2 m
0.0 0.0
(a) Time (hr) Cost: 0.0652 $/m3 (b) Time (hr) Cost: 0.0651 $/m3
Off-Peak Peak Off-Peak Peak
5.0 5.0 300
)m Off-Peak Upper Trigger Level: 5 m )m
( le v e L re34 .. 00 Off-Peak Lo Pw eae kr T Ur pig pg ee rr
T
L re igv ge el:
r
4 L. e4
v
m
el : 2.2 m
( le v e L re34. .0 0 12 84 00 )s /L ( w o
lF
ta ta p
W2.0 W2.0 120 m
k n k n u P
a a
T1.0 T1.0 60
Peak Lower Trigger Level: 1.2 m
Pump Flow (VSP)
0.0 0.0 0
(c) Time (hr) Cost: 0.0649 $/m3 (d) Time (hr) Cost: 0.0625 $/m3
Figure 4.4: Daily tank level variation of the one-pipe network: cost optimal solutions for pump operating (a) Control
Case B; (b) Control Case C; (c) Control Case D; (d) Control Case E
Control Case C: Cost Minimization
For Control Case C, the combination of trigger levels and scheduling, the cost was reduced slightly
compared with the previous control cases at $0.0651/m3. Due to the high demands at the end of the peak
period, shutting the pump down during this time would not be feasible. Therefore, an additional decision
variable in the form of a pump startup during the peak time was considered as well as those proposed in
the methodology. The time range for this pump startup was 4 to 10 p.m. at an increment of 5 min, which
allowed the tank level to stay above 1 m, and a pump shutoff was considered between 10 and 11:30 p.m.,
also at an increment of 5 min. The optimal cost solution found using this strategy again had wide trigger
levels of 1 and 5 m, the pump was started again at 5:35 a.m. and this allowed the tank to fill exactly for
the start of the peak period [Figure 4.4(b)]. During the peak period, the optimal solution started the pump
at 6:20 p.m. and then shut it down at 10:20 p.m. to have the tank empty at the end of the peak period.
Control Case D: Cost Minimization
Using variable trigger levels in Control Case D found an optimal solution that maintained a low water level
during the peak period, with trigger levels of 1.2 and 2.2 m, and a high water level during the off-peak
period, with trigger levels of 4.4 and 5.0 m [Figure 4.4(c)]. Even though this solution had a slightly greater
percentage of pumping during the peak period compared with the Control Case C solution, it reduced the
static head for much of the simulation period and was therefore slightly cheaper at $0.0649/m3.
35 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
Control Case E: Cost and GHG Minimization
Scheduling in Control Case E was able to find solutions with reduced cost and GHG emissions compared
with the other control cases. The best cost solution using VSPs used lower pump speeds throughout the
off-peak period to fill the tank exactly at the start of the peak period [Figure 4.4(d)] and had a cost of
$0.0625/m3. The use of FSPs was more expensive than VSPs; the cost optimal solution using FSP had
a cost of $0.0656/m3. FSP scheduling was less flexible than VSP operation and was not able to completely
fill the tank for the start of the peak period. The optimal solution for GHG emissions pumped constantly
throughout the day at reduced speeds, compared with the cost optimal solution, which pumped as much
as possible in the off-peak period. This resulted in a cost of $0.0682/m3 and GHG emissions of 0.2156
kgCO -eq/m3, both of which are lower than for all of the solutions (cost or GHG optimal) presented in
2
Table 4.3 for Control Case A.
4.4.2 Case Study 2: South Australian Network
The second case study was a real-life WDS in South Australia, consisting of 324 pipes, 278 nodes, two
pumps (one on standby), one reservoir, and two tanks (Figure 4.5). This case study was chosen to show
the advantages and disadvantages of the different pump operating control cases and objectives for a real
network. With only one pump operating, the comparison between the control cases could be made clearly
and their effect on the objectives more easily understood. With an average daily peak day demand of 30.7
L/s compared with the pump operational flow of 126 L/s, the pump in this network was oversized and only
required to operate for 8 h each day. Under the current operational regime using trigger levels of 3.96 and
5.54 m, almost half of this pumping occurred during the peak electricity tariff period (Figure 4.6), when
electricity rates were much higher (22 c/kWh compared with 9 c/kWh for off-peak). Cost and GHG
emissions for the current operation were $0.0360/m3 and 0.1460 kgCO -eq/m3, respectively. The
2
maximum tank water level was 7.92 m, with a minimum tank water level set at 2.5 m, representing 30%
of the full volume to account for emergency reserves and dead storage. Trigger level values considered
in the optimization ranged from 4.0 to 7.9 m at an increment of 0.1 m, with the initial tank water level set
at 4.0 m for all simulations. The minimum pump speed multiplier was calculated to be 0.87 [Eq. (4.4) with
a pump shutoff head of 92 m and maximum static head of 69.4 m], so choices for multipliers ranged from
0.88 to 1.0 in 0.04 increments (Table 4.2). The optimization results for all control cases for this network
are presented in Tables 4.4 and 4.5 and discussed in the following sections.
Control Case A: Cost and GHG Minimization
For Control Case A, the optimal trigger levels to minimize cost for this network were 4.0 and 6.1 m, costing
$0.0219/m3, 39% less than the current operation (Table 4.4). The pumping in this solution occurred
entirely within the off-peak period, with the tank filling between the hours of 11 p.m. and 6:30 a.m. and
then draining for the rest of the day [Figure 4.7(a)]. Optimizing for GHG emissions found that trigger levels
of 4.0 and 4.3 m reduced emissions to 0.1434 kgCO -eq/m3, a 1.8% saving on the current operation
2
(Table 4.4).
Control Cases B, C, and D: Cost Minimization
With all pumping able to be completed in the off-peak period, the addition of a reduced upper trigger
(Control Case B) found optimal solutions with the same cost as the optimal trigger levels solution (Control
Case A). Regardless of switch time, the optimal upper trigger level was greater than 6.1 m (the optimal
upper trigger level value for Control Case A), and the reduced upper trigger level varied such that all the
pumping could still be achieved during the off-peak period. This indicated that it was better to pump
entirely within the off-peak period with the ultimate upper trigger level in effect rather than pump
throughout the day with a reduced static head. Control Cases C and D, which also attempted to take
36 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
advantage of the off-peak tariff and reduce the static head during the peak period, were also not useful
(Table 4.4). In Control Case C, the optimal scheduled pump start occurred at times when the pump was
already on and the optimal pump stop when the pump was already off, leaving the operation to be entirely
governed by the trigger levels, which were the same as for Control Case A. In Control Case D, the
operation was governed by the off-peak lower trigger level and the peak upper trigger level, which were
the same as the Case A optimal trigger levels.
EL 106 m
EL 100 m
EL 162 m
Figure 4.5: South Australian Network
Off-Peak Peak
8.0 160
)m
(
le
v
e67 ..0
0
11 24 00 )s
/L
(
w
L5.0 100 o
re
ta4.0 80
lF
p
m
W3.0 60 u
k P
n a2.0 40
T
1.0 20
0.0 0
Time (hr) Cost: 0.0360 $/m3
Figure 4.6: Daily tank level (solid) and pump flow (dashed) variation for the South Australian network: current
operation
Control Case E: Cost and GHG Minimization
Optimization of VSP scheduling (Control Case E) found a marginally better solution to the cost optimal
trigger levels operation with a cost of $0.0218/m3. It pumped at a reduced speed from 11 p.m. to 6 a.m.
and then at full speed for the last hour of the off-peak period [Figure 4.7(c)]. While the reduced speed
would lead to less friction loss through the system and hence reduced energy requirements, there was
an extra 90 min of pumping that meant the cost and GHG emissions from the VSP solution were very
similar to the trigger levels solution (Table 4.4). The optimal GHG solution pumped during half of the time
37 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
periods, including during the middle of the day when the emissions factors were lowest. This solution had
emissions of 0.1419 kgCO -eq/m3, a reduction of 2.9% compared with current operation.
2
Table 4.4: Optimal solutions for each pump operating control case for the South Australian network
Control Cost Cost Diff. GHGs (kg GHG Diff. Peak Energy Off-Peak
Objective
Case ($/m3) (%)a CO 2-eq/m3) (%)a (%) Energy (%)
A Cost 0.0219 -39.2 0.1466 +0.4 0.0 100.0
A GHGs 0.0438 +21.6 0.1434 -1.8 71.3 28.7
B Cost 0.0219 -39.2 0.1464 +0.3 0.0 100.0
C Cost 0.0219 -39.2 0.1466 +0.4 0.0 100.0
D Cost 0.0219 -39.2 0.1466 +0.4 0.0 100.0
E Cost 0.0218 -39.5 0.1459 -0.1 0.0 100.0
E GHGs 0.0466 +29.3 0.1419 -2.9 80.4 19.6
aA negative difference indicates that the cost or GHGs in the optimal solution is less than the current operation (cost:
$0.0360/m3, GHG: 0.1460 kg CO -eq/m3).
2
Table 4.5: Optimal solutions for each pump operating control case for the South Australian network with a smaller
pump
Control Cost Cost Diff. GHGs (kg GHG Diff. Peak Energy Off-Peak
Objective
Case ($/m3) (%)a CO 2-eq/m3) (%)a (%) Energy (%)
A Cost 0.0291 -19.2 0.1339 -8.3 31.0 69.0
A GHGs 0.0385 +7.0 0.1320 -9.6 64.7 35.3
B Cost 0.0291 -19.3 0.1339 -8.3 31.0 69.0
C Cost 0.0291 -19.2 0.1339 -8.3 31.0 69.0
D Cost 0.0291 -19.3 0.1139 -8.3 31.0 69.0
E Cost 0.0280 -22.3 0.1348 -7.7 27.0 73.0
E GHGs 0.0409 +13.4 0.1315 -10.0 72.6 27.4
aA negative difference indicates that the cost or GHGs in the optimal solution is less than the current operation (cost:
$0.0360/m3, GHG: 0.1460 kg CO 2-eq/m3).
Replacement with a Smaller Pump
In order to apply all of the pump operating control cases to a real-life network, the current pump was
assumed to be replaced with a smaller pump that would be required to pump for more than the 8 off-peak
hours each day. The current pump operated at a flow of 126 L/s at a head of approximately 70 m. Because
the average demand was 30.7 L/s, a pump with a flow of approximately 40 L/s at a head of 70 m was
selected. This pump required roughly 13 h of pumping per day. The shutoff head was 80 m, which gave
a minimum pump speed multiplier of 0.93 and thus multipliers between 0.94 and 1.0 in increments of 0.02
were considered.
Control Case A: Cost and GHG Minimization with a Smaller Pump. Using the smaller pump in Control
Case A, the optimal trigger levels for cost were 4 and 5.5 m; at $0.0291/m3, this was more expensive than
with the original pump (Table 4.5). This suggests that when there are large differences between the peak
and off-peak cost of electricity, it may be more economical to install a larger, more expensive pump but
have reduced operating costs by only pumping during the off-peak period. With a smaller pump, the tank
did not fill as quickly and hence some of the pumping occurred during the peak period [Figure 4.7(b)].
This solution still reduced the cost by 19% compared with the cost of the current operation with the original
pump (Table 4.5). Using the smaller pump reduced both GHG emissions and cost at the same time. The
cost-optimal solution for Control Case A with the original pump slightly increased GHG emissions
compared with the current operation. With the smaller pump, however, the cost-optimal trigger levels also
reduced GHG emissions by approximately 8%. The optimal GHG trigger levels when the smaller pump
was used were 4.0 and 4.7 m, further apart than with the original pump.
38 |
ADE | Publication 1: Comparison of Pumping Regimes for Water Distribution Systems to Minimize Cost and
Greenhouse Gases
Off-Peak Peak Off-Peak Peak
8.0 8.0
)m7.0 )m7.0
(
le6.0
Upper Trigger Level: 6.1 m (
le6.0
Upper Trigger Level: 5.5 m
v v
e e
L5.0 L5.0
re
ta W4.0
Lower Trigger Level: 4.0 m
re
ta W4.0
Lower Trigger Level: 4.0 m
3.0 3.0
k k
n a2.0 n a2.0
T T
1.0 1.0
0.0 0.0
(a) Time (hr) Cost: 0.0219 $/m3 (b) Time (hr) Cost: 0.0291 $/m3
Off-Peak Peak Off-Peak Peak
8.0 160 8.0 160
)m
( le
v
e67 .. 00 11 24 00 )s
/L
( w
)m
( le
v
e67 .. 00 11 24 00 )s
/L
( w
L5.0 100 o L5.0 100 o
re
ta
W
k34 .. 00 68 00
lF
p
m
u
P
re
ta
W
k34 .. 00 68 00
lF
p
m
u
P
n a2.0 40 n a2.0 40
T T
1.0 20 1.0 20
0.0 0 0.0 0
(c) (d)
Time (hr) Cost: 0.0218 $/m3 Time (hr) Cost: 0.0280 $/m3
Figure 4.7: Daily tank level and pump flow variation for the South Australian network: cost optimal solutions for (a)
Control Case A with original pump; (b) Control Case A with smaller pump; (c) Control Case E with original pump; (d)
Control Case E with smaller pump
Control Cases B, C, and D: Cost Minimization with a Smaller Pump. With the use of the smaller pump,
Control Cases B, C, and D found optimal solutions that had effectively the same operation as for the
Control Case A solution (Table 4.5). With a reduced upper trigger level (Control Case B), the ultimate
upper trigger level was ineffective and the pump was entirely controlled by the reduced upper trigger level
at an optimal level of 5.5 m. When trigger levels and scheduling were combined (Control Case C), the
same optimal trigger levels were found and the scheduled pump startup occurred when the pump was
already on, and similarly the pump shut down when the pump was already off. With variable trigger levels
(Control Case D), the peak levels governed the operation; during the off-peak period, the tank level did
not reach the off-peak upper trigger level, and the peak upper trigger level, at 5.5 m, controlled when the
pump stopped.
Control Case E: Cost and GHG Minimization with a Smaller Pump. VSP scheduling (Control Case E)
with the smaller pump gave a better result than the trigger level operation with a cost of $0.0280/m3 (Table
4.5); however, it was still more expensive than with the original pump because some pumping in the peak
period was required [Figure 4.7(d)]. The optimal GHG pump schedule with the smaller pump provided the
best GHG solution for all of the South Australian network solutions in Tables 4.4 and 4.5 with emissions
of 0.1315 kgCO -eq/m3 giving a 10% saving on the current operation.
2
4.5 Conclusions
A single-objective genetic algorithm model has been developed to optimize pumping operations in water
distribution systems. It was combined with a new toolkit for EPANET2 that allowed optimization of more
complex pump operating strategies than have previously been considered to be performed. Five different
pump operating control cases were implemented, using various types of trigger levels, scheduling, and
39 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Abstract
Water security has become an increasing concern for many water system managers due to climate
change and increased population. In order to improve the security of supply, alternative sources such as
harvested stormwater, recycled wastewater and desalination are becoming more commonly used. This
brings about the need for tools to analyze and optimize systems that use such sources, which are
generally more complex than traditional water systems. Previous methodologies have been limited in their
scope and cannot be applied to all types of water sources and systems. The framework presented in this
paper has been developed for holistic analysis and optimization of water supply and distribution systems
that use alternative water sources. It includes both design and operational decision variables, water and
energy infrastructure, simulation of systems, analysis of constraints and objectives, as well as policies
and regulations which may affect any of these factors. This framework will allow users to develop a
comprehensive analysis and/or optimization of their water supply system, taking into account multiple
types of water sources and consumers, the effect of their own design and operational decisions, and the
impact of government policies and different energy supply options. Two case study systems illustrate the
application of the framework; the first case study is a harvested stormwater system that is used to
demonstrate the importance of simulation and analysis prior to optimization, the second utilizes four
different water sources to increase security of supply and was optimized to reduce pump energy use.
44 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
5.1 Introduction
A changing climate and increasing population have put a strain on traditional water resources, which
typically rely on natural catchment water. This has made water security an increasing concern for many
water system managers, who have investigated options for reducing demand and supplementing supply.
Alternative water sources, such as harvested stormwater, recycled wastewater and desalination, are
increasingly being used to improve water security of cities and towns. Methods for simulation, analysis
and optimization of traditional potable water distribution systems (WDSs) cannot necessarily be directly
transferred to systems that use alternative water sources. Therefore there is a need to develop a
methodology specifically for alternative water source systems, which includes both hydraulic and
hydrologic considerations, as well as the many additional parameters and variables associated with
alternative water sources. There are many modelling tools used in current practice for integrated water
management, such as eWater Source, WEAP (Water Evaluation and Planning System) and Mike Basin.
These modelling tools do not include hydraulic simulation, and therefore may not accurately represent
performance of urban water networks. Moreover, this framework is not software, rather its purpose is to
guide water system managers in how to best simulate and optimize their systems, particularly those that
integrate multiple water sources, and natural and human-made systems. The framework should be used
to determine which system components need to be modelled, which type of modelling tools are most
appropriate, what regulations and policies need to be taken into account and how to evaluate the
performance of the system.
The framework introduced in this paper can be applied to the optimization of the design and operation of
water supply and distribution systems from source to consumer, considering multiple traditional and
alternative sources, multiple uses and multiple objectives. Electrical energy sources and their effect on
electricity prices and greenhouse gas (GHG) emissions are included, as are several types of government
policies that may affect the design, operation, data and evaluation of the system. The objectives of this
paper are to (1) develop a generalized framework that could be applied to any water supply and/or
distribution system optimization problem and (2) outline the application of this framework to two case
study systems with a focus on optimizing their operation.
5.2 Literature Review
Since 2000, there has been significant consideration of the concept of water security (Cook and Bakker,
2012) as water is increasingly seen as a fundamental and finite resource (Bogardi et al., 2012).
Consequently, the use of alternative sources, such as harvested stormwater, desalination, recycled
wastewater and rainwater, has gained traction (Fielding et al., 2015). Harvested stormwater schemes are
often decentralized and used for non-potable supply such as household gardening and irrigation of public
reserves (Naylor et al. 2012), however, in some cases are also used for potable supply (McArdle et al.,
2011). While desalination is a climate independent (and therefore more reliable) source, is often not the
most cost effective or environmentally sensitive option (Becker et al., 2010; Miller et al., 2015). Recycled
wastewater is also climate independent, and generally used for large scale non-potable applications
(Muga and Mihelcic, 2008; Oron et al., 2014), however, it can also be used for indirect or direct potable
supply (Rodriguez et al. 2009; Nagal 2015). Domestic rainwater tanks are increasing in popularity and
have benefits of reducing water usage from utilities and reducing stormwater runoff from houses
(Campisano and Modic, 2012, Umapathi et al., 2013). Demand management strategies have also been
used to reduce per capita consumption and therefore reduce the pressure on limited water supplies
(Dawadi and Ahmad, 2013; Friedman et al., 2014).
45 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Some alternative sources, such as harvested stormwater, introduce additional complexity to the problem
of modeling and optimization than has been previously considered for traditional water systems (Marchi
et al., 2016a). There is, for example, the need to consider the supply and distribution systems together,
rather than separately, as it is less likely that there will be large storages isolating the supply side from
the distribution side. When including the supply side, longer simulation times often need to be used,
requiring rainfall and evaporation scenarios to be taken into account. The security of supply with regard
to climate change needs to be considered (Paton et al., 2014; Cai et al., 2015), as some sources are
climate dependent and some are climate independent. The social acceptability of using particular sources
for particular applications and the willingness of consumers to pay more for alternative source systems to
be constructed and maintained may need to be incorporated (Hwang et al., 2006; Londoño Cadavid and
Ando, 2013; Fielding et al., 2015). The perception of risks associated with alternative water source
systems by water system managers may also present a barrier to the implementation and success of
such systems (Dobbie and Brown 2012; West et al. 2016). Many alternative sources also have associated
externalities that result in either cost or benefit to the user, such as reduced effluent flow to the ocean or
receiving water body by reusing wastewater and reduced urban stream flows by harvesting stormwater
(Marchi et al., 2016a).
The increased use of alternative water sources then raises the question of how such systems should by
analyzed and optimized to ensure they are implemented as effectively as possible. Stokes et al. (2014)
developed a framework for optimizing the cost and GHG emissions of WDSs, taking into account both the
design and operation of the system, energy sources and GHG emission factors. This study, however, was
applicable only to traditional WDSs, with no consideration of the supply side and alternative water sources.
Chung et al. (2008) developed a mathematical model for evaluating integrated water supply systems with
decentralized treatments. Multiple sources, uses, transportation and treatment systems can be
considered, however only surface water, groundwater and recycled wastewater sources are included.
This model does not incorporate any optimization procedure, only analysis of different options developed
by the user. Makropoulos et al. (2008), with further developments in Rozos and Makropoulos (2013),
produced a decision-support tool for modeling the urban water system from source to tap. The software
can be used to select combinations of water saving strategies and technologies, including how much
water from each type of demand (for example domestic, commercial) is obtained from each source and
how the system is operated. It uses a demand-oriented, water balance approach and does not include
capability for other types of simulation models such as hydraulic and hydrologic modeling.
Uncertainty, particularly with regard to climate change, is an important consideration that has been taken
into account in several methodologies. Paton et al. (2014) developed a framework for water supply system
planning with alternative sources and climate change considerations, while Beh et al. (2014, 2015)
developed two methods for optimal sequencing of urban water supply augmentation options under deep
uncertainty regarding demands and climate. The research by both Paton et al. (2014) and Beh et al.
(2014, 2015) considered only the planning of water supply projects, and did not optimize the specific
design or operation of the systems. Sequencing is also considered in Cai et al. (2015), however, in this
case it is applied to planning of drought mitigation strategies in agricultural systems. They consider
multiple decision stages in which options such as infiltration ponds, parallel terraces, irrigation triggering
threshold and irrigation water sources can be implemented. Marchi et al. (2016a) developed a
methodology for optimizing the design of harvested stormwater systems taking into account future climate
scenarios; however, it does not apply to other types of alternative sources or optimization of system
operation. It does include a detailed analysis of the associated externalities, such as reduced peak flows
and improved economic value of properties near stormwater schemes. Ashbolt et al. (2014) introduced a
46 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
framework for planning of short-term operations for water systems using surface water, groundwater,
desalination, and recycled wastewater with multiple objectives and multiple inflow replicates to account
for uncertainty. Long-term operating strategies and the design of the system were not included and the
operating strategies considered were limited to bulk water transfers and not the operation of pumps and
smaller storages.
5.3 Framework for the Optimization of Alternative Water Sources
The framework presented in the current paper was developed to guide the modeling and optimization of
water supply and distribution systems that use alternative water sources. It is comprised of several
components and sub-components that fit within an optimization structure, for example, a multi-objective
evolutionary algorithm (Figure 5.1). The options component [OPT] describes the potential ‘decision
variables’ that are available in an optimization problem, that is, the factors that can be changed in order
to produce a different outcome. This includes both the initial design of the water supply and distribution
infrastructure and the long- and short-term rules that govern the operation of the system once it has been
commissioned. The infrastructure component [INF] describes the physical components of the system that
need to be modeled and the data associated with each, including both water infrastructure and energy
infrastructure, which may affect the evaluation of electrical energy cost and life-cycle GHG emissions.
There is also a government policy component [G] that covers the policies from regulating bodies that may
affect other aspects of the framework. The analysis component [ANL] describes the simulation of each
potential system configuration and evaluation against objectives and constraints. The optimization
algorithm [OA] investigates different possible combinations of decision variables from the options
component, models the system according to the infrastructure component and evaluates it using the
analysis component to find the optimal solution(s).
Details of the components and sub-components are shown in Figure 5.1 and described in Sections 5.3.1
to 5.3.4. Table 5.1 summarizes the parameters that need to be considered in the optimization and
simulation of alternative water source systems with respect to the different items that are presented in
Figure 5.1 and in the following sections. There are three (non-exclusive) categories that each parameter
may be placed in – decision variables, parameters that are set, and uncertain parameters. Decision
variables are parameters that the user may be able to examine using optimization. It is important to note
that in most optimization problems, not all of these parameters will be available as decision variables at
once, and it is likely that only a small number will be considered. For example, when optimizing pump
operations for an irrigation system, only the first three ‘decision variables’ shown in Table 5.1 (pump
schedules, tank trigger levels, and demand scheduling) may be considered. The remaining parameters
that are designated as decision variables in Table 5.1, particularly those relating to the design of the
system (for example, delivery system layout and pump sizing) would already be set and not able to be
optimized if the existing infrastructure cannot be modified. The parameters that are set are those that very
rarely, if ever, are able to be optimized by the user. These include parameters that would be controlled by
external sources, for example consumers of domestic or commercial demands, pipe manufacturers and
higher level government and regulatory bodies; and also parameters that need to be predefined to a
known or assumed value before optimization or simulation can be performed, for example, fire
demand/reserve, hydrologic/hydraulic variables and objective and constraint selection and definition. The
final category, uncertainty, designates those externally set or predefined variables that are not well known
or may be subject to change in the future and therefore may need to be considered in a sensitivity analysis.
While the selected values of decision variables have an impact on the performance of a system, they are
generally within the control of the decision maker, and therefore are not classed as ‘uncertain’. It is
important to note that the categorization in this table is indented as an indication of how each parameter
47 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
is typically treated. There are, of course, exceptions to this, as almost all of the parameters could be
considered as decision variables if desired and have some associated uncertainty. For example,
environmental flows have been designated as an externally set parameter, as it is likely that the operator
of a system will have to meet requirements set by an external organization such as the Environmental
Protection Agency. They may, however, want to investigate providing greater environmental flows, or
show the benefits of reducing their environmental flow requirements and being able to supply more water
elsewhere.
5.3.1 Options Component [OPT]
The options component covers the potential decision variables (and the range of possible choices for the
decision variables) for an optimization problem. This component is split into two sub-components; the
operational decisions sub-component [O] and the design decisions sub-component [D]. Design decisions
include elements that can be changed before a system is constructed, such as the layout and capacities,
materials and other properties of the various infrastructure components. Operational decisions include
elements that can be changed after construction during the daily management of the system, such as the
operating rules for pumps and valves and allocation of water from different sources.
Operational Decisions Sub-Component [O]
Both short- and long-term operations are considered in the operational decisions sub-component. The
critical aspects of this sub-component (items in bold can be optimized), as shown in Figure 5.1 and Table
5.1 are:
[O1] the specific short term operating strategies including pump schedules (when pumps are
turned on or off based on time), trigger levels (water levels in tanks or other storages that
determine when pumps or valves turn on or off), irrigation or demand schedules (for
systems where they can be pre-determined), valve settings and operating rules, and
pressure settings for pumps (to maintain the set pressure at a particular point).
[O2] the specific long term operating strategies including volumetric allocation of water from
different alternative sources, trigger levels (for example in reservoirs) that determine
allocations from different sources or water demand restriction levels, switch times between
different operating regimes (for example between different trigger level sets for different
seasons) and power source selection.
[O3] the overall short-term operating strategy, including operating rules that are optimized in [O1]
and operating rules that are pre-set and are not to be optimized (acting as constraints).
Where there are multiple operating rules, the priority of each rule and order they are enforced
in is important to consider.
[O4] the overall long-term operating strategy, including operating rules that are optimized in [O2]
and operating rules that are pre-set and are not to be optimized. Again, the priority and order
of the rules is important to consider.
Most systems have multiple operating conditions to meet and therefore multiple operating rules will be in
place. Prioritization of the different operating rules is important, and this may be set by the operator or be
chosen by the optimization tool. This component requires information from the government policy sub-
component ([G] in Figure 5.1), specifically in terms of water source licensing and environmental flow
regulations. These policies would typically be regulated by local or state government departments or the
environmental protection authority. Operational rules set in this sub-component will inform the simulation
sub-component [S] as they will need to be represented in any simulation model(s) of the system.
48 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Table 5.1: Summary of parameters for the design and operation of alternative water source systems
Parameter Decision Parameter Uncertain Relevant Items in
Variable* that is set Parameter Figure 5.1
OPERATIONAL INPUTS [O]
Pump schedule X O1
Tank trigger levels X O1
Tank / storage maximum and minimum allowable X O1, W3, W11
levels
Demand pattern (irrigation, agriculture) X O1, D4, W13
Demand pattern (domestic, commercial, industrial) X X O1, D4, W13
Demand flow rate (peak, average, peak day) X X O1, D4, W14
Valve settings or operating rules X O1
Pump pressure settings X O1
Volumetric allocation of water X O2
Reservoir trigger levels X O2
Switch time between operating regimes X O2
Priority ranking of operating rules X O3, O4
DESIGN INPUTS [D] AND WATER INFRASTRUCTURE [W]
Water source selection X D1, W2
Water source infrastructure (layout, capacity) X D1, W2
Treatment type selection X D2, W8
Treatment infrastructure (layout, capacity, treatment X D2, W8
rate/level)
Delivery system type selection X D3
Delivery system layout (lengths, elevations, junctions, X D3, W7, W10,
tank locations) W12, W15
Pipe material and diameters X D3, W7, W10,
W12
Pipe parameters (unit cost, pipe wall roughness (ε), X X (ε) D3, W6, W7,
wall thickness, embodied energy) W10, W12
Pump sizing X D3, W5, W9
Pump performance characteristics and cost X D3, W4
Tank sizing (capacity, height, diameter) X D3, W3, W11
Fire demand / reserve X D3, W11
Water user type selection X D4
Rainfall / streamflow series X X W1
Reservoir capacity and volume curve X W3
Pond (e.g. wetland) capacity and volume curve X W3
Prioritization rules for demands types X W15
OTHER INPUTS [P], [G] AND [S]
Power source selection X X P1, P3, G5
Electricity tariff structure and cost X X P2
GHG emission factors X X P3, G5
Fit-for-purpose requirements X G1
Water license amounts X G2
Environmental flow amounts X G3
Discount rate X X G4
Hydrologic variables (e.g. permeability) X S1
Hydraulic variables (e.g. water temperature) X S3
OPTIMIZATION PROBLEM FORMULATION [E]
Objective selection X E1
Objective function(s) X E2
Constraint selection X E3
Constraint limits (maximum and minimum) X E4
Penalty costs X E4
*Note: Parameters specified as decision variables are shown in bold throughout Sections 5.3.1 and 5.3.2.
50 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Design Decisions Sub-Component [D]
This sub-component incorporates all of the design decisions that are available to the designer for the
entire water supply and distribution system, from source to user. The critical aspects of this sub-
component (items in bold can be optimized), as shown in Figure 5.1 and Table 5.1 are:
[D1] the water sources selected to be used including natural catchments, harvested stormwater,
recycled wastewater, groundwater, imported water, domestic rainwater, desalination,
domestic greywater and sewer mining; and the layout and capacity of source
infrastructure.
[D2] the types of treatment selected including centralized treatment at plants such as
mechanical filtration, chemical dosing, ultraviolet treatment and ozonation, and decentralized
in situ treatments such as gross pollutant traps, wetlands and biofilters; and the layout,
capacity, dosing rates and retention times for treatment facilities.
[D3] the type and configuration of the delivery system used including potable, non-potable
(for example dual reticulation systems to deliver recycled water), centralized and
decentralized, and the infrastructure design variables such as system layout, pipe sizes,
lengths and materials, pump sizing, valve sizing, and tank sizing.
[D4] the types of water users that are supplied by the system including potable, irrigation,
agriculture, industrial, non-potable domestic/commercial and firefighting, and the demand
rate and pattern for water use (for example, scheduling of irrigation demands).
Regulations on fit-for-purpose water use from the government policy component [G] in Figure 5.1 inform
what water sources can be used for particular applications and these are likely to be specified by state or
federal government departments or health authorities. Generally, sources such as harvested stormwater
and recycled wastewater cannot be used for potable supply and rather serve non-potable demands in
dual-reticulation systems or are supplied to irrigation, agricultural and industrial users. There may be some
systems, however, in which necessary approvals have been obtained to use these sources for potable
supply. The design decisions are inputs to the water system infrastructure sub-component [W] which
describes the system elements and data to be modeled.
5.3.2 Infrastructure Component [INF]
The purpose of this component is to describe the infrastructure that needs to be modeled in order to
evaluate the objectives and constraints of the problem. There are two sub-components; the water system
infrastructure sub-component [W] and the electrical energy infrastructure sub-component [P]. Water
system infrastructure includes the specific aspects of the water supply and distribution system and the
data required, including construction and maintenance costs. Electrical energy infrastructure includes the
power source (fossil fuel types and renewable types) and the electricity price and GHG emission factor
data needed.
Water System Infrastructure Sub-Component [W]
This sub-component includes the specific infrastructure aspects of the water system design and the
relevant data that is needed in order to simulate it. Most systems and optimization problems will not require
all of these factors to be considered or modeled; however, the purpose of this framework is to cover a
large range of the possible requirements for an optimization and hence the scope is intentionally broad.
The water system infrastructure sub-component [W] as shown in Figure 5.1 represents a system with one
water source, one treatment plant, one storage tank and one demand node. In reality, many systems will
have more than one of each of these components, particularly the treated storage [W11] and demand
51 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
node [W15]. Pumping of water between storages may occur in multiple stages, particularly when there is
a large difference in elevation. For typical centralized potable WDSs, all treatment will occur at one water
treatment plant. In decentralized systems such as for harvested stormwater schemes, however, treatment
may occur in multiple stages. For example, a gross pollutant trap may be located on an urban creek before
the water is collected in a harvest pond, then the water may be pumped to be treated through a wetland,
and then treated again in a treatment plant.
The critical aspects of this sub-component (items in bold can be optimized) as shown in Figure 5.1 and
Table 5.1 are:
[W1] the rainfall or inflow scenarios for the water source; for example rainfall or streamflow
scenarios for natural catchments and stormwater sources, or a sewer system flow pattern
for recycled wastewater. Sources such as desalination and, depending on the temporal scale
of the optimization, groundwater, do not usually require an inflow scenario. Rainfall and
streamflow scenarios may be a data series obtained from measurements at gauging stations
or modeled in a hydrologic simulation program [S1]. Multiple inflow scenarios may be used,
particularly for systems with highly variable inflows. Losses such as evaporation and
infiltration may also need to be taken into account for sources with large open storages such
as reservoirs and natural water ways.
[W2] the source type as described in [D1] with input from [W1].
[W3] the raw water storage; this may be a reservoir (typical for a natural catchment), a harvest
pond for a stormwater system, a tank (for example for a recycled wastewater system) or an
aquifer for groundwater. Associated data including capacity, a volume curve, elevation,
height and diameter is required.
[W4] characteristics of available pumps such as performance curves (head, efficiency, and power
against flow), cost, rated speed and variable speed pump (VSP) information where
applicable.
[W5] the pump transferring water from the raw water storage to a treatment facility, requiring data
from [W4].
[W6] pipe size and material information such as available diameters, unit costs, pipe wall
roughness, wall thickness and embodied energy. For new pipes, this information will be
easily obtained from the pipe manufacturer. For existing systems, however, there may be
some uncertainty in these parameters if detailed records of the ‘as constructed’ system and
any pipe replacements have not been kept. In addition to this, the pipe wall roughness of
existing pipes will generally be uncertain. Pipe wall roughness increase as pipes age, and
pipe condition assessment may be needed to provide an estimate.
[W7] the pipe system transferring water from the raw water storage to the treatment facility, pipe
lengths and layouts need to be known as well as information from [W6].
[W8] the treatment facility that will produce water of the required quality based on the source
type and demand type. Characteristics of the individual treatment methods as described in
[D2] need to be known.
[W9] the pump transferring water to a treated storage, requiring the same data as [W5].
[W10] the pipe system transferring water to a treated storage, with the same information as [W7]
required.
[W11] the treated storage, for example, a tank or multiple tanks that are typically at a high elevation
point of the network in order to supply demands by gravity. Data required includes the
elevation, height, diameter and maximum and minimum allowable water levels.
52 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
[W12] the pipe system transferring water from the treated storage to consumers, which again
requires information as in [W7]. This pipe system is likely to be more complex than those in
[W7] and [W10], particularly for systems with many different demand nodes. For systems
with only one source of water, [W7] and [W10] are likely to be single pipelines. For
decentralized systems with only one specific consumer, [W12] will also most likely be a
single pipeline. Most systems, however, have much more than one demand point and as
such distribution systems are often looped or branched systems that require more complex
analysis than single pipelines.
[W13] demand scenarios that will be applied to the demand nodes, consisting of a pattern of
demand multipliers over a day, week or year. There may be multiple demand scenarios
required for a system, for example, if there are different types of demand nodes (such as
domestic, commercial, industrial) or different seasonal demands.
[W14] the peak demand is the demand rate that is typically used to size the system components
and so will affect the design of the system. The demand scenarios [W13] are more likely to
affect the operation of the system as the demand varies over the simulation time. The peak
day demand (average demand over the peak day), the peak hour demand (the average
demand over the hour with maximum consumption in the peak day) and average demand
rates may also be required. Fire loading demands and other emergency conditions will affect
the design of the system, for example storage tanks should be sized to be able to provide
demand in the case of fires, other emergencies and system failures (e.g. if the supply to the
tank is cut off).
[W15] the demand nodes for the consumers, these may be different types of users as described in
[D4] and require information from [W13] and [W14]. Different types of users will have different
demand rates [W14] and demand patterns [W13]. When simulating the system, an average
demand rate will often be used with the demand pattern, rather than the peak demand.
Systems with multiple demand nodes may prioritize different types of demands over other,
for example, irrigation systems using non-potable water may prioritize high profile sport fields
over reserves with no formal usage.
Choices made in the optimization of the design decisions sub-component [D] in Figure 5.1 will be inputs
to the water system infrastructure sub-component. There may be other parameters that are not decision
variables in the optimization (as differentiated in Table 5.1) though are still required by this sub-component
in order to simulate the system. The construction and maintenance costs of each of the infrastructure
components needs to be known in order to calculate the initial construction cost and ongoing costs as
part of life-cycle economic costing. Information collected through this sub-component will be input to the
simulation sub-component [S] depending on the types of simulation models used and to the evaluation
sub-component [E] through the construction cost or other factors calculated for the specific objectives of
a problem.
Electrical Energy Infrastructure Sub-Component [P]
The electrical energy infrastructure sub-component includes any power infrastructure that affects the
electricity prices and GHG emission factors. The critical aspects of this sub-component as shown in Figure
5.1 and Table 5.1 are:
[P1] the breakdown of power sources including fossil fuel sources such as coal and oil, and
renewable sources such as solar, wind and hydrothermal.
53 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
[P2] the electricity price tariff structure, which may be a peak and off-peak structure, or multi-part
(more than two price levels) and could include a peak demand charge which applies to the
peak electricity power usage in each month.
[P3] the GHG emission factor, which is based on the power source breakdown [P1] and may vary
with time, either in the short-term (with sources that do not have storage such as solar panels
and wind turbines) or the long-term (as fossil fuel sources tend to be phased out and
renewable sources become more popular).
Climate and energy policy [G5] in the government policy component in Figure 5.1 will affect the power
source breakdown and electrical energy pricing now and into the future. This is likely to be regulated by
a federal government department or body. Information from this sub-component is used to calculate
electrical energy costs in order to evaluate life-cycle economic costs and also to calculate life-cycle GHG
emissions in the evaluation sub-component [E].
5.3.3 Government Policy Component [G]
The government policy component covers policies by regulating bodies at any level (local, state, federal)
that may affect other aspects of the framework. These policies need to be considered over the operational
life-span of the system, for example, climate and energy policy may affect future energy sources and
therefore affect future GHG emissions. The critical aspects of this component as shown in Figure 5.1 and
Table 5.1 are:
[G1] fit-for-purpose water use, which may be regulated by state or federal governments or health
agencies and affects which water sources [D1] and water uses [D4] can be combined in the
design decisions sub-component. It may also guide which design decisions (for example,
treatment) are appropriate.
[G2] water source licenses, which may be regulated by local or state governments or the
environmental protection agency, depending on the catchment size, and will affect the
amount of water available from particular sources for allocation in long-term operations [O4].
[G3] environmental flows, which similarly to water source licenses may be regulated by local or
state bodies depending on the size of the catchment and affect the amount of water available
for allocations [O4].
[G4] the discount rate applied to operational costs and GHG emissions in life-cycle analysis [E1].
This is unlikely to be set by a government body and rather will be informed from outside the
decision making team, generally by recommendations from economists.
[G5] climate and energy policy set by state and federal governments will affect the energy sources
available now and in the future, therefore affecting GHG emission factors and any GHG
objectives [P].
5.3.4 Analysis Component [ANL]
The analysis component uses information from the options, infrastructure and government policy
components to simulate the system and evaluate how it performs relative to the objectives and
constraints. Within an optimization algorithm, the analysis component is used to assess multiple
combinations of decision variables from the options component to determine how they perform. There are
two sub-components within the analysis component; the simulation sub-component [S] and the evaluation
sub-component [E]. The simulation sub-component includes the modeling aspects of the problem and the
key variables that are required to be output from the models in order to evaluate the system. Optimization
objectives and constraints are defined in the evaluation sub-component, which also provides information
to the optimization algorithm as to which of the potential solutions perform best.
54 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Simulation Sub-Component [S]
The simulation sub-component considers the type of simulation model that is most applicable to the
particular system and problem, and specifies the key variables that need to be calculated in the model(s).
The critical aspects of this sub-component as shown in Figure 5.1 and Table 5.1 are:
[S1] the hydrologic simulator, which is required if rainfall scenarios need to be transformed to
streamflow, typically for systems using natural catchment water or harvested stormwater.
[S2] the mass balance model, which may be required for systems that have multiple water
sources with long-term allocation decisions, particularly if there are different rainfall and
evaporation scenarios to be considered for the storages.
[S3] the WDS hydraulic simulator, which is required to analyze pump and pipe systems that
transfer water between different storages and treatments and to consumers.
[S4] information on constraints, such as yield from a hydrologic model, environmental releases
and system reliability from a mass balance model, and nodal pressures, pipe velocities,
pump switches and tank levels from a hydraulic model.
[S5] the water levels in storages, which are important particularly when considering operational
decisions, such as trigger levels, and for constraints, such as system reliability.
[S6] the power usage from any pumps or treatment facilities, which are important in informing the
ongoing electrical energy costs as part of life-cycle economic costing. Generally a WDS
hydraulic simulator is required to model detailed pump operations and therefore accurately
estimate the pump power usage.
Each of the three types of models will require different simplifications or assumptions depending on the
particular system. For example, mass balance modeling will generally only consider one pump operating
point so may not accurately calculate the pump power usage. When deciding which type of model to use
for a particular problem, the user will need to consider the different simplifications, assumptions,
advantages and disadvantages of each model. Trade-offs between accuracy of outputs and simulation
run times need to be considered. For example, when optimizing both short- and long-term operations of
a system, there is likely to be a trade-off between using a hydraulic simulator for detailed hydraulic
information and using a mass balance model for shorter run times. Most problems may ideally use
elements from more than one type of model; however, using multiple models will increase computational
complexity and run times. Wherever possible, the most applicable type of model should be selected and
augmented with the required elements from other types of models. Depending on the particular system
and optimization problem, there may be other key variables that need to be calculated in the simulation
models. For optimization of pumping operations, which is the focus of the case studies in this paper,
storage water levels and pump power usage are the most important. Existing hydrologic, mass balance
and hydraulic simulators, for example, MUSIC, WATHNET and EPANET, have often been used in
conjunction with optimization algorithms and should be taken advantage of where possible rather than
creating individual simulators for different problems.
Information from the operation decisions sub-component [O] will be input to the simulation sub-component
as the overall operating strategy for the system ([O3] and [O4]) will need to be modeled. Short-term
operations are likely to be considered in a hydraulic simulator and long-term operations, including
allocations, in a mass balance model. Parameter data on the physical components of the system from the
water system infrastructure sub-component [W] are also required as inputs for this sub-component.
Constraint information is provided to the evaluation sub-component to compare the systems performance
against specified requirements. Energy usage is used to calculate objective functions such as life-cycle
55 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
economic costs and life-cycle GHG emissions. Simulating systems prior to optimization is an important
step to help inform the formulation of the optimization problem and provide a check that results from the
optimization are reasonable.
Evaluation Sub-Component [E]
The purpose of the evaluation sub-component is to compare the performance of each of the potential
solutions to the objectives and constraints of the problem. The critical aspects of this sub-component as
shown in Figure 5.1 and Table 5.1 are:
[E1] the specific objective(s) to be considered in the optimization; typically, minimizing life-cycle
economic cost is a primary objective (or a component of that such as construction cost or
operational cost individually). Other possible objectives include minimizing spills from
reservoirs and other storages, minimizing life-cycle GHG emissions (or a component of that
such as embodied energy from construction or operational emissions), minimizing
supplemental potable water supply (in systems using non-potable sources), maximizing
water quality, maximizing reliability and minimizing environmental impact.
[E2] the objective function(s) to be optimized; multiple objectives may be evaluated as individual
functions in a multi-objective optimization algorithm or combined into a single function for
use in a single objective optimization algorithm. It is important to consider how each objective
should be formulated, for example, when optimizing short-term pump operations to minimize
ongoing costs, the objective function may be evaluated in terms of cost per volume of water
pumped, as this will take into account the amount of water delivered to consumers. Reliability
of a system may be formulated in different ways, for example minimizing the time spent with
water restrictions applied or minimizing the time spent below a certain storage level. Some
objectives may be more difficult to quantify, such as minimizing environmental impact, so
more specific objectives may be required, for example, maximizing environmental flow or
minimizing the change in a water body’s natural hydrological regime. Simplifications and
assumptions may be required to formulate some objectives as mathematical functions.
When performing multi-objective optimization, trade-offs between the different objectives
should be considered by the development of Pareto fronts, allowing the decision maker to
determine which Pareto optimal solution best fits their needs (see examples in Wu et al.
2010a, 2010b, 2012a, 2012b, 2013).
[E3] the specific constraints to be considered as described in [S4].
[E4] the evaluation of the constraints compared to the limits set by the user; maximum and/or
minimum values for each constraint need to be specified. Some constraints may be flexible,
for example, if an environmental flow is set by a regulator, the operator could consider
increasing the set environmental flow as a decision variable in the optimization. There are
several different ways constraints can be incorporated into the optimization algorithm.
Penalty functions are often used for single-objective problems. They add value (often a
monetary amount) to the objective function in a minimization problem and remove value from
the objective function in a maximization problem based on the magnitude of the constraint
violation, therefore making solutions that violate constraints less desirable (Nicklow et al.,
2010). Care must be taken when formulating penalty functions to keep solutions that only
slightly violate constraints in consideration during the optimization process, while ensuring
the feasibility of the final optimal solutions. For multi-objective problems, a constraint-
handling technique that will ensure feasible solutions are retained in preference to infeasible
56 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
solution is often employed. An example of this is the constraint tournament selection
procedure introduced by Deb et al. (2002).
Information about the objectives is received from the simulation sub-component [S] and from the
calculation of construction, maintenance and electrical energy costs based on the water system
infrastructure sub-component [W] and simulation sub-component. A discount rate for costs or GHG
emissions may be set in the government policy sub-component [G] which will impact the ongoing costs
and emissions in a life-cycle analysis. The discount rate may be informed by economists, such as the
Stern review which recommends low discount rates for projects that lead to the production of GHG
emissions (Stern 2006). Information about the performance of each potential solution in relation to the
objectives and constraints is provided to the optimization algorithm in order to find the best solutions.
5.3.5 Optimization Algorithm [OA]
The optimization algorithm is used to determine which solution(s), out of many potential solutions to the
problem, performs best in relation to the objective function(s). The procedure used to set up the
optimization will depend on the type of algorithm chosen; however, the first step is generally to define the
decision variables, objectives and constraints of the problem. This will then guide what aspects of the
system need to be modeled and what data is required in order to take into account all of the decision
variables and that will provide information for all of the objectives and constraints. Multiple potential
solutions to the problem form the ‘solution space’ and the optimization algorithm guides the search of this
solution space towards the global optimum. The size of the solution space depends on the number of
decision variables and number of choices available for those decision variables. More complex problems
are often described as having a more ‘rugged’ solution space, meaning the optimization algorithm is more
likely to get trapped in local optima and will have more difficulty finding the global optimum. When a single
objective optimization algorithm is used, one optimal solution will be found, while in multi-objective
optimization, a Pareto front will be developed with multiple solutions representing different trade-offs
between the objectives.
Most optimization algorithms have parameters that need to be defined by the user, such as the number
of generations or iterations and the population size in evolutionary algorithms. Although the choice of
these parameters does not influence the components shown in Figure 5.1, they have an effect on the
optimal solutions found by the algorithm. In general, the most effective set of parameter values to use
will vary between different optimization problems and the size of the solution space can only give some
indication of what parameter values to use. In fact, multiple parameter sets should be tested in order to
find the most appropriate values for the specific problem. Ideally, the chosen parameter set should find
the same optimal solution regardless of the starting point or initial solution(s) for the optimization. Dandy
et al. (1996) presented an improved genetic algorithm formulation for optimization of WDS design. Five
different parameter sets were trialed on both their improved genetic algorithm and a comparatively simple
genetic algorithm. They acknowledged that parameter selection does require some judgement on the part
of the user, however, they found their optimization results to be relatively insensitive to the parameter
choice, particularly for the improved genetic algorithm. As well as the effect of various parameter values,
different optimization algorithms will be more suited to different problems. This issue has been addressed
by the development of hybrid algorithms, such as AMALGAM (a multi-algorithm, genetically adaptive
multiobjective approach proposed by Vrugt and Robinson (2007)), which combines several different
optimization algorithms to improve search efficiency. These hybrid algorithms also have the benefit of
requiring little to no parameter specification by the user.
57 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
5.3.6 Sensitivity Analysis
As identified in Table 5.1, values of some input parameters (for example, describing the network or water
demand loadings) are uncertain or subject to change in the future. Sensitivity analysis can be performed
to account for a wide range of possible future conditions when optimizing and simulating systems.
Variation of a particular parameter may result in different Pareto fronts (in multi-objective optimization) or
different optimal solutions (in single objective optimization), as seen in Wu et al. (2010b) when they
considered variations in discount rates. These various Pareto fronts or optimal solutions along with the
various parameter values that produced them can then be provided to the decision maker. Sensitivity
analysis will also help to identify if there are any uncertain parameters that do not affect the optimal results.
Robustness of the optimized solutions can also be explored a-posteriori: in general, solutions that perform
well for many different possible conditions are more desirable from the decision makers’ point of view.
Climate change provides an additional source of uncertainty for the parameters identified in Table 5.1 –
detailed discussion of this is omitted from Sections 5.3.6.1 to 5.3.6.4 as it is covered in Section 5.3.6.5.
Demand
In some applications, such as irrigation and agriculture, the demand rate and pattern may be deterministic
[O1], either the water supplier has control over the consumption, or may be able to work with those who
do to determine an optimal demand schedule. For other applications, such as domestic, commercial and
industrial, the demand rate and pattern depends on the consumption of water by multiple individual users
[D4, W13, W14, W15], and therefore has greater uncertainty. Historical consumption can provide some
level of assurance as to how water may be used in the future, at least on an aggregated scale. Diurnal,
weekly and seasonal demand variations need to be considered. In the future, factors such as climate
change, population growth and water saving initiatives will affect how water is consumed and therefore
impact demand rates and patterns. Emergency conditions and system failure are by their nature
unpredictable and this should be taken into account when designing and operating WDSs.
An example of how demand uncertainty can be considered in the optimization of WDS design is the study
by Basupi and Kapelan (2015). The demand at each time step was based on a normal distribution with a
gradually increasing mean (based on deterministic demand forecasts) and an increasing standard
deviation. Monte Carlo or Latin Hypercube simulation was included in their methodology to consider
multiple demand scenarios. Each solution in the Pareto front was also further analyzed against three
demand projections – average, optimistic (low overall demand) and pessimistic (high overall demand).
Their results demonstrated the value of flexible WDS design over deterministic approaches when
considering uncertainty.
Rainfall and Streamflow
Rainfall and streamflow inputs [W1] may be required for systems using natural catchment water,
harvested stormwater or imported water, and they are often treated with higher uncertainty than demands
(Seifi and Hipel, 2001; Reis et al., 2005). Within the current climate, there may be multiple realizations of
possible rainfall and streamflow series (for example dry or wet years). Beh et al. (2015) considered rainfall,
as well as population and temperature, as uncertain variables in their optimal sequencing methodology
for water supply system augmentation. They considered both climate and hydrologic variability: seven
possible future climate scenarios provided different forecasted rainfall reductions, and within each of these
seven scenarios, 20 stochastic replicates of the rainfall sequence were produced. Different Pareto fronts
were produced for each climate scenario, with the more severe scenarios finding solutions that required
greater system augmentation and therefore had higher costs and GHG emissions. The robustness of
58 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
each Pareto solution was calculated based on the average reliability and vulnerability of the solution over
the 20 rainfall sequences for the particular climate scenario.
Electricity and GHG Emissions
Power source(s) [P1], electricity tariffs and costs [P2] and GHG emission factors [P3] will generally be
known for the present time, however, it may not be clear how they will change in the future. The mix of
power sources changes naturally over time, as different power plants are built or decommissioned. This
change in power source types over time, as well as technical advancements will affect the cost and GHG
emissions associated with electrical energy generation. The electricity market and economic factors will
also affect the cost of electrical energy over time. Changes in electricity and GHG emissions can be an
important factor to consider during an optimization problem, as shown in the following examples. Blinco
et al. (2014) studied the optimization of pump operations in WDSs in relation to the minimization of GHG
emissions and the use of different power source scenarios, showing that optimal tank trigger levels can
be influenced by the variation in emission factors. Wu et al. (2012a) considered three different electricity
tariff scenarios, which increased over time, and three different GHG emission factor scenarios, which
decreased over time, in the optimization of WDS design. The different electricity tariff and emission factor
scenarios affected the solutions found in the Pareto front and their overall costs and GHG emissions.
Discount Rate
A discount rate [G4] may be used in life-cycle analysis for both ongoing economic costs and ongoing
GHG emissions. In practice, discount rates on economic costs vary significantly between different
organizations, generally from 2% to 10% (Rambaud and Torrecillas, 2005), while many water utilities use
discount rates in the range of 6% to 8% (Wu et al. 2010a). When selecting discount rates, consideration
should be given to whether both economic costs and GHG emissions are discounted, if they have the
same discount rate, and if intergenerational equity is taken into account using social discount rates.
Various social discount rates have been proposed for discounting ongoing costs; the Intergovernmental
Panel on Climate Change (IPCC) adopted a zero discount rate over a period of 100 years, after which no
consideration for future costs or benefits is given (Fearnside 2002), other suggestions include 1.4% (Stern
2006) for projects that are impacted by climate change, 2-4% (Weitzman, 2007) and a time declining rate
(Gollier and Weitzman, 2010). Wu et al. (2010b) gave an example of a sensitivity analysis of discount
rates in the optimization of WDS design for minimization of costs and GHG emissions. Discount rates of
0%, 1.4%, 2%, 4%, 6%, 8% and a time declining rate were applied to the economic costs, with GHG
emissions either not discounted at all, or discounted at the same rate as costs. They found that the
different discount rate scenarios produced different Pareto fronts; when GHG emissions were discounted,
the solutions tended to have lower capital costs and higher operating emissions.
Climate Change
Management of water resources in the developed world has been based on an assumption of stationarity
– that is, ‘that natural systems fluctuate within an unchanging envelope of variability’ (Milly et al. 2008).
The effects of human-induced climate change make this assumption no longer valid (Milly et al. 2008),
and introduce additional sources of uncertainty for many parameters. Uncertainty introduced by climate
change is twofold – firstly, the impacts of climate change increase the uncertainty of future weather
conditions; and secondly, our response to the threat of climate change and the types of adaption methods
that will be utilized in the future are uncertain. Climate change affects the magnitude and temporal and
spatial distribution of rainfall, temperature and other environmental factors, thus the possible rainfall and
streamflow series to consider for the future will likely be different to the present. Changes to temperature
and other environmental factors will also affect the hydrology of natural and urban catchments and
59 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
therefore change how rainfall will transform to runoff or streamflow. Climate change impacts will also affect
how people consume water, for example, higher temperatures and lower rainfall may drive people to water
their gardens more. In order to simulate future climate conditions, general circulation models (GCMs) are
often used in conjunction with future emissions scenarios. According to Mpelasoka and Chiew (2009),
‘GCMs are the best tools available for simulating global and regional climate systems’, however, the
information provided is generally too coarse for applications to catchment runoff, and therefore some kind
of downscaling is required. The modeling uncertainty of both the GCMs and downscaling methods
increases the uncertainty of future climate scenarios (Paton et al., 2013). In 2000, the IPCC introduced
several emissions scenarios (termed SRES scenarios) projecting future global GHG emissions. The
various scenarios are based on different assumptions of the mix of energy generating technologies (fossil
fuel or non-fossil fuel dominant) and population, economic and technological growth (IPCC 2007).
The extent to which we can reduce our GHG emissions will affect the magnitude of climate change
impacts on rainfall and temperature. With the growing concerns of climate change and sustainability,
renewable sources such as solar and wind will become more prevalent and replace fossil fuel sources
such as coal and gas. This may affect electricity pricing and GHG emissions from power generation.
Multiple future power source scenarios assuming different levels of climate change mitigation may need
to be considered. Other climate change adaption strategies include economic incentives such as carbon
taxes and cap and trade systems, which may affect economic analysis of WDSs. As discussed in Section
3.6.4, when climate change and intergenerational equity are considered, social discount rates of 0%,
1.4%, 2-4% and time declining rates have been proposed.
Paton et al. (2013) analyzed the sources of uncertainty relating to climate change and their impact on
water supply security. They considered 19 different scenarios with different combinations of six SRES
scenarios, seven GCMs and six demand projections, as well as 1000 stochastic rainfall replicates. They
found that the impact of the different sources of uncertainty on the optimal solutions varied over the 40-
year planning period, with some having a greater effect in the short-term and others a greater effect in the
long-term. Roshani and Filion (2014) investigated the impact that different climate change abatement
strategies have on water main rehabilitation. They consider six carbon abatement strategies with different
combinations of two discount rates (1.4% and 8%) and three carbon tax scenarios (no tax, ‘fast and deep’,
and ‘slow and shallow’). Using a low discount rate and implementing a carbon tax encouraged the
optimization algorithm to find solutions that invested in rehabilitation early, to reduce the cost of continuing
leaks, pipe repair, energy use and GHG emissions.
5.4 Case Studies
The utility of the framework described in the previous sections will now be explored by two different case
studies that have different water sources and many variables that need to be considered. These case
studies are provided as an example of how the framework could be applied to optimize system operations.
The first case study is a managed aquifer recharge (MAR) system that harvests stormwater from an urban
creek for irrigation of reserves and sporting fields. This case study demonstrates the importance of
analyzing the system by simulation prior to optimization in order to formulate the optimization problem.
The second case study is a water supply system in a rural town that supplies potable water from multiple
alternative water sources. This system is optimized for minimization of energy use of the many pumps
used to transfer water from the various sources.
5.4.1 Ridge Park Managed Aquifer Recharge – Case Study 1
Ridge Park is located in the Adelaide metropolitan area in South Australia, within the City of Unley local
government area. The scheme supplies harvested stormwater to sports fields and recreational reserves
60 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
For existing systems, simulation analysis of the current operation is an important first step in formulating
the optimization problem. Results of current operational simulations can highlight areas for improvement
that can then be focused on in the optimization. The operation of the Ridge Park stormwater harvesting
system was split between winter and summer operations and both were simulated in EPANET to
determine current pump operational costs. Trigger levels (related to volumes in the three storages as
shown in Table 5.2) control when the pumps in the Winter Harvesting and Injection system turn on and
off (Table 5.2). The Bore Pump is also controlled by trigger levels in the Storage Tank. During summer,
Pump 3 is controlled by the irrigation demands, which are on a schedule so that different reserves are
irrigated on different nights (Table 5.3). Pump 3 is a VSP and is operated at 80% of full speed for injection
(such that the flow is less than the 7 L/s maximum for injection) and 75% of full speed for irrigation (such
that the target pressure downstream of the pump is achieved at the expected demand rates). Both
systems were simulated for a period of one week in EPANET, with a 15 minute hydraulic time step and
five minute reporting time step. Several week-long streamflow series for the available flow in Glen
Osmond Creek at a daily resolution were used in the harvesting and injection model (Figure 5.4). A
peak/off-peak electricity price tariff applied to the entire system; a peak price of 25.53 c/kWh was applied
from 7am to 9pm on weekdays, and an off-peak price of 15.26 c/kWh was applied over night and on
weekends (tariff pattern and simulations started on a Sunday).
Table 5.2: Trigger levels for the Ridge Park System
Current Setpoint
Storage and Trigger Level Type Start Pump Stop Pump
Volume (%) Level (m)
Harvest Pond High Level 80 1.6 1 -
Harvest Pond Low Level 50 1.0 - 1
Biofiltration Basin High Level 90 0.80 2 1
Biofiltration Basin Low Level 50 0.59 - 2
Storage Tank High Level 90 2.25 3 2, Bore
Storage Tank Low Level 70 1.75 Bore 3
Table 5.3: Irrigation demand schedule for the Ridge Park System
Reserve Demand Rate (L/s) Duration/day (hr) Start Time Irrigation Days
Ridge Park 1 3.53 8.33 9:30 PM Mon & Wed
Ridge Park 2 3.53 8.67 9:30 PM Tues & Thurs
Fraser Reserve 1.41 5.83 9:30 PM Mon & Wed
Ferguson Ave Reserve 2.00 5.00 9:30 PM Tues & Thurs
Scammell Reserve 2.15 6.00 10:00 PM Tues & Thurs
Fullarton Park 1 3.85 1.67 10:00 PM Mon & Wed
Fullarton Park 2 3.85 6.67 10:00 PM Tues & Thurs
Fern Ave Reserve 3.53 3.33 10:00 PM Mon & Wed
Windsor St Reserve 2.20 8.00 8:30 PM Tues & Thurs
Henry Codd Reserve 1.10 8.00 10:00 PM Mon & Wed
Unley Oval 5.57 9.00 9:00 PM Sun, Mon & Wed
Winter Harvesting and Injection System current pumping operation results
When there was adequate water available, such as in Streamflow Series 1, 4 and 5, the volume of water
injected into the aquifer (by Pump 3) was a little over 3 ML per week (Table 5.4). This was significantly
less than the volume available, which reflects the limited flow rate of Pump 3 (7 L/s for injection to the
aquifer), as well as the water that would be lost to overflow when the inflow rate is greater than the flow
rate of Pump 1 (approximately 22 L/s). The total pump energy cost estimate for the harvesting and
injection system ranged from $163 to $267 per week, with an average of $235 per week. Pump 1 was the
most cost-effective to run, while Pump 3 was the most expensive. Pumps 1 and 2 operated at similar
62 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
times throughout the day, however, Pump 2 has much lower efficiencies, which increased its energy use.
Pump 3 operated at a lower flow rate but much higher head than Pumps 1 and 2, and was more likely to
be switched on for the entire day, which contributed to its higher cost of operation. Pumps 1 and 2 turned
on and off very frequently, and operated at a much higher flow rate than Pump 3 (Figure 5.5). The flow
rate of Pump 3 in Figure 5.5(c) reduced over the week as the headloss through the bore increased from
assumed clogging of the bore. As the storages are relatively small, in particular the storage tank, it did
not take long for them to be filled and emptied (Figure 5.6), which contributed to the frequent pump
switches. The current trigger levels in the Storage Tank are very close together (70% and 90% volume)
as a result of possible pump priming issues that occurred during the commissioning of the system. These
close together trigger levels also contributed to the short fill and empty times.
70.0
)s
/L 60.0
(
w
o 50.0
lfm
a
e 40.0
r
t
S
e 30.0
lb
a
lia
20.0
v
A
10.0
0.0
1 2 3 4 5 6 7
Day
Series 1 Series 2 Series 3 Series 4 Series 5
Figure 5.4: Streamflow series used for simulation of the Winter Harvesting and Injection operation
Table 5.4: Current operation results for the Winter Harvesting and Injection System
Streamflow Available Volume Cost (c/kL) Volume Injected Total Cost
Series (ML/wk) Pump 1 Pump 2 Pump 3 (ML/wk) ($/wk)
1 19.0 0.64 2.28 5.49 3.14 267
2 2.29 0.68 2.32 6.19 1.76 163
3 6.19 0.69 2.23 5.87 2.44 222
4 15.4 0.64 2.24 5.46 3.18 258
5 29.7 0.63 2.25 5.47 3.16 264
Average 14.5 0.66 2.26 5.70 2.74 235
Summer Extraction and Irrigation System current pumping operation results
Simulation of the irrigation system gave a total weekly pump energy cost of $90 (Table 5.5). The Bore
Pump was more expensive overall, however, cost less per megaliter than Pump 3. This occurred because
while the Bore Pump has a greater efficiency than Pump 3, it also has a higher flow and head, which
increased the energy consumption. The higher volume pumped from the bore contributed to a lower cost
rate than Pump 3. All of the pumping for this system occurred overnight (Figure 5.7) when irrigation of all
fields is allowed. The Bore Pump turned on and off very frequently when it was operating, again due to
the small capacity of the Storage Tank which meant it did not take long for the pump to fill the operating
volume (Figure 5.8).
63 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Table 5.5: Current operation results for the Summer Extraction and Irrigation System
Pump Volume (ML/wk) Cost (c/kL)
Bore Pump 1.93 3.52
Pump 3 0.57 3.97
Total $90.3 / week
20
16
)s 12
/L
(
w
o 8
lF
4
0
Simulation Time (hr)
Total Irrigation Demand Gravity Irrigation Demand Pressure Irrigation Demand and Pump 3 Flow
Figure 5.7: Current demand rate and pump flows for the Irrigation System
2.5 30
25
2.0
)m
(
le 1.5
20
)s
v
e L
r e 1.0
15
/L
( w
o
ta 10 lF
w
W
0.5
5
0.0 0
Simulation Time (hr)
Storage Tank Bore Pump
Figure 5.8: Storage Tank level and Bore Pump flow for the Summer Extraction and Irrigation System
Optimization Formulation
Initially, optimization of the Ridge Park system was considered to be an operational problem, however,
results of the current operation simulation suggest that design decision variables need to be considered
as well. Replacing Pumps 1 and 2 with models that would operate at much lower flow rates (to reduce the
headlosses) and increasing the size of the Storage Tank will be considered along with operational
decision variables (Table 5.6). These design decisions would aim to counter-act mismatched pump rates
(Pumps 1 and 2 operating at a much higher rate than Pump 3) and small storage volumes that lead to
frequent pump switches. Short-term operational decisions include trigger levels in the Harvest Pond,
Bioretention Basin and Storage Tank that will govern when pumps are turned on and off, a schedule for
irrigation (that is, which reserves will be irrigated at which times), and VSP multipliers for Pump 3. In the
65 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
current operation, VSP multipliers for Pump 3 were selected to ensure the required flow rate (for injection)
and pressure (for irrigation) were achieved. With different levels in the Storage Tank considered, the VSP
multipliers for Pump 3 can be altered, especially if efficiency is improved. If the pump priming issues
discussed earlier were to be resolved, trigger levels that utilize the full height of the Storage Tank (rather
than the 20% range in water elevation that is currently used) would be considered in the optimization.
There are also long-term decision variables deciding when to switch between summer and winter
operation and vice versa (Table 5.6). As the scheme injects to and extracts from the aquifer through the
same bore, it is not possible to frequently switch between injecting and extracting water, therefore there
will be only two switch times per year; one going into winter operation and one going into summer
operation. The decision variables presented in Table 5.6 may all be considered together in an optimization
problem, however, they could also be analyzed prior to optimization in a simulation sensitivity analysis.
Simulating the system initially with the different pump models and storage tank sizes could help to decide
if these actions are worthwhile considering in an optimization formulation. Engineering judgement may be
sufficient to determine which pump model(s) would be best to replace Pumps 1 and 2, and therefore
reduce the size of the optimization problem.
Table 5.6: Possible decision variables for the Ridge Park MAR Scheme
SHORT-TERM WINTER HARVESTING AND INJECTION OPERATION
Pump 1 Off Harvest Pond Level Low
Bioretention Basin Level High
Pump 1 On Harvest Pond Level High
Pump 2 Off Bioretention Basin Level Low
Storage Tank Level High
Pump 2 On Bioretention Basin Level High
Pump 3 Off Storage Tank Level Low
Pump 3 On Storage Tank Level High
Pump 3 Speed Storage Tank Level
SHORT-TERM SUMMER EXTRACTION AND IRRIGATION OPERATION
Bore Pump Off Storage Tank Level High
Bore Pump On Storage Tank Level Low
Irrigation Schedule Days of Irrigation at each Reserve
Start Time of Irrigation at each Reserve
Pump 3 Speed Required Demand Rate
LONG-TERM OPERATIONS
Day to Switch Between Seasonal Summer to Winter
Operational Regimes Winter to Summer
SYSTEM DESIGN
Storage Tank Size Doubled, 5 times, 10 times current size
Pumps 1 and 2 Selection of pump curves with lower
operating rates
Constraints on the system include an environmental flow for Glen Osmond Creek, an extraction limit from
the Aquifer and meeting the weekly irrigation volumes for each reserve in the summer (Table 5.7). If there
was not enough water harvested over the winter to supply the summer demands, a potable back-up
supply is available at a cost. The main objective for this case study is to minimize the pump energy cost;
there is also a secondary objective of minimizing the number of pump switches. To create an incentive
for the optimization to find solutions that harvest more water, the cost objective includes the energy cost
for the harvesting and distribution operation as well as the cost of purchasing potable water if the
harvested volume is not enough to supply demand. The objective function is formulated as the cost per
volume of water harvested as another means to ensure enough water is harvested from the system during
winter to supply summer irrigation. During the conceptualization and design of this scheme, regulations
66 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
alternative sources to fill up Suma Park reservoir just before a rainfall event that would supply water from
the natural catchment at no cost or energy output. As this system supplies potable demands, it is
undesirable to apply water restrictions to consumers, thus minimizing time spent in restrictions is
important. Objectives for the perceived environmental impact and water quality will be formulated as a
preference ranking between the different sources based on community views of which sources are better
for the environment and water quality. The constraints of the problem include environmental flows for the
Macquarie River (downstream of the pumping station offtake point) and stormwater schemes, a water
source license for the Macquarie River and extraction limits on the groundwater bores (Table 5.8).
Macquarie Pumps Blackmans
River Pumps S1a, b Creek Pond
M1a, b
Pumps M2a, b Ploughmans Stormwater
Holding
Scheme
Spring Creek Pumps Pond
Brooklands
Catchment Balancing S2a, b, c
Wetland
Storage M1
Spring Creek
Dam Pumps
Pumps S5a, b
M3a, b Batch Somerset
Balancing
Ponds Wetland
Storage M2
Suma Park Suma Park Pumps S6a, b
Mitchell
Catchment Dam
Wetland
Pumps S3a, b, c
Balancing Pump
Bore 5 Pump G3b
Storage G1 G1b
Pumps S4a, b
Balancing
Escort
Pump G1a Storage G3
Pump Wetland
Shearing
G2b Pump G3a
Shed Bore
Balancing Showground Cargo
Storage G2 Bore Wetland
Pump G2a
Figure 5.9: Orange Integrated Supply System process schematic – inflow to Suma Park Dam
Ploughmans Clifton Bores Pump M1
Wetlands Stormwater Ponds 182 ML 177 L/s, 317 m
264 ML
35 ML Pump M2
186 L/s, 190 m
Blackmans Pond
Maquarie River
40 ML
> 300000 ML/yr
Showground Bore
280 ML Pump M3
186 L/s, 188 m
Suma Park Dam
19 000 ML
Orange
Pump
Pond
Figure 5.10: Orange Integrated Supply System layout and data
68 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
Table 5.8: Possible constraints for the Orange Integrated Supply System
Constraint Value
Macquarie River Environmental Flow > 108 ML/day
Blackmans’ Creek Environmental Flow > 20 ML/day
Ploughmans’ Creek Environmental Flow from Pump S4 > 0.4 ML/day
Ploughmans’ Creek Environmental Flow from Pump S5 > 2 ML/day
Ploughmans’ Creek Environmental Flow from Pump S6 > 2 ML/day
Clifton Grove Aquifer Extraction < 182 ML/year
Showground Aquifer Extraction < 280 ML/year
Macquarie River Extraction License < 12 ML/day
Energy Optimization Formulation
In this section, the developed framework is applied to the Orange Case Study to help set up the
optimization procedure and identify the components and data to be modeled. Note that the model has
been built taking into account all possible objectives of the system, however, the example of results
presented here will focus on the minimization of energy consumption.
As all components of the system have already been constructed and considered sufficient for the
operation of the system, there are no design decisions to consider, only operating decisions. For this case
study, operating decisions consist of trigger levels in the various storages. These types of decision
variables are chosen considering the control system available at each pump station (based on storage
levels and not on time of the day) and the fact that the controls have to be defined for an operational
horizon of one year or longer. As all of the pump stations have two or more pumps arranged in parallel,
having different trigger level values may have a large impact on the operating point of the pumps and
consequently their energy consumption. It is also likely that different trigger levels will be chosen for peak
and off-peak electricity tariff periods when they are included in a cost optimization. For this system a
peak/off-peak electricity tariff applies on weekdays, with weekends priced at the off-peak rate. A peak
monthly electrical energy demand charge also applies to the Macquarie River pipeline pumping system.
In order to assess the performance of different tank trigger levels, the infrastructure to be modeled
includes the natural and urban catchments for the surface water and stormwater systems respectively,
Suma Park reservoir, pipelines and pumps in the groundwater, Macquarie River and stormwater systems,
and wetlands and storage ponds in the stormwater systems.
In general, the system could be modelled using hydrologic models, mass balance models, and/or
hydraulic models. The choice of which model(s) will be used depends on the objectives and the processes
to be modelled, on the available data and the computational times. In particular, hydrologic modeling is
usually used to transform rainfall to runoff for the natural and urban catchments. For this case study,
inflows inputs or approximate relationships between rain and flows were provided by previous studies by
the Orange City Council. Hydraulic models are usually used for short term operations: pump energy costs
can be computed accurately based on the hydraulic equations. Mass balance modeling is usually used
for assessing the system in long term operations, as it can quickly compute the water available after
evaporation and other losses in the system have occurred and after minimum environmental flows have
been released. It cannot, however, take into account the non-linearity in the hydraulic equations and
therefore assumptions need to be made in regard to the flow delivered by the pumps in the system. While
hydraulic simulation would be most appropriate for the pumping stations in the system as they have
multiple pumps and sometimes have connected pipelines, mass balance models would need to be used
to compute the additional processes, such as evaporation and the release of minimum environmental
flows that need to be taken into account given the long duration of the simulation. During an optimization
process, simulating each potential solution using both a mass balance and a hydraulic model would
69 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
increase considerably the computational time, particularly if data transfer between the two models was
required. It is therefore suggested that the primary simulation tool should be a hydraulic solver. Rainfall-
runoff modeling could be performed pre-optimization, and supplemental code added to a hydraulic model
to account for functionality of a mass balance model. This would allow for consideration of the evaporation
from and rainfall directly to reservoirs, changes to demands based on water restrictions and environmental
flows that depend on the combined volume of two reservoirs (Spring Creek and Suma Park), infiltration
losses when transferring water between reservoirs and peak power demand charges.
Another important issue to consider is what simulation time step should be used. Using a shorter time
step will increase the accuracy of this hydraulic analysis and often results in feasible optimization times
for storages that empty or fill in a day or two (as would likely be the case for the stormwater ponds and
Macquarie pipeline balancing storages). Simulating the behavior of Suma Park dam is more challenging,
however, as the variations in the water levels can have a period of several years. Thus, the computation
times with a short time step become prohibitively long. A balance needs to be found between using a
short enough time step for the detailed hydraulics and a long simulation time for the large storages without
having a prohibitively large computational time. Given the data availability (there is 118 years of rainfall
and inflow data available, with a daily time step) the time step chosen is one day.
Given that the time-step is automatically shortened by the hydraulic solver chosen (EPANET in this case),
the model of the real system has been simplified in order to avoid excessive computational times. In
particular, given that the levels in the balancing storages along the Macquarie pipeline vary rapidly, these
storages were removed and the pipeline simulated with two parallel pumps, each representing the
equivalent of the three stages of pumping (that is, the pump curves for Pumps M1a and b in Figure 5.9
were adjusted such that they represented Pumps M2a, M3a and Pumps M2b, M3b as well). This
simplification is considered acceptable as the pumps in series in the Macquarie pipeline will usually be
operated at the same time, given that each pump will still be controlled also by the level of Suma Park
Dam. Longer computational times were also caused by the small storages after the groundwater bores.
The pumps used for extraction from the aquifers (Pumps G1a, G2a and G3a in Figure 5.9) operate at
relatively consistent rates, and as such they could be removed from the model and their energy use
accounted for relative to the volume pumped from the second pump in each system (Pumps G1b, G2b
and G3b respectively). To take into account the limited volume available from the groundwater bores, the
storage tanks in the groundwater system each had a volume equivalent to a year’s allocation for the
respective bores. All of the stormwater pumps except for Pump S2c and Pump S3c, which are standby
pumps and not in use, were included in the model. As well as the operating point of the pumps changing
depending on the number of pumps used in parallel, there may be slight differences in efficiency and
therefore energy use, and thus including all pumps here provided more accuracy.
All of the pumps included in the model were controlled using rule-based controls in EPANET, with
conditions based on levels in one or more storages as well as time. Conditions based on downstream
storages were considered as decision variables, while conditions based on upstream storages were fixed
(Table 5.9). For the Macquarie pumps, there were also conditions based on the flow in the river to ensure
that no water would be taken when there was not enough water available. There were four possible
decision variables for each pump, a lower and upper trigger level in both the peak and off-peak time. For
optimization of energy use, only two are required, as peak and off-peak tariffs are not considered. As the
model was set up for other objectives including cost, which does use a peak and off-peak electricity tariff,
the capability to choose different trigger levels in different periods was incorporated. A maximum of 15
pump switches per day per pump were allowed, and the end level of Suma Park Dam was constrained to
70 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
16 m (to be approximately the same as the start level). Based on license conditions, Macquarie River
water can only be used when the Suma Park Dam level is below 90%, so choices for Pump M1a and M1b
trigger levels in Suma Park Dam are more restricted than for other pumps.
Table 5.9: Decision variables and fixed controls for the Orange Integrated Supply System
Pump Station Action Storage(s) Controlling Operation Decision Variable or Fixed
Macquarie Pump M1a, M1b Off Suma Park Dam Level High Decision Variable
Macquarie Pump M1a, M1b On Suma Park Dam Level Low Decision Variable
Stormwater Pump S1a, S1b Off Holding Pond Level High Decision Variable
Blackmans Stormwater Pond Level Low Fixed
Stormwater Pump S1a, S1b On Holding Pond Level Low Decision Variable
Blackmans Stormwater Pond Level High Fixed
Stormwater Pump S2a, S2b Off Batch Ponds Level High Decision Variable
Holding Pond Level Low Fixed
Stormwater Pump S2a, S2b On Batch Ponds Level Low Decision Variable
Holding Pond Level High Fixed
Stormwater Pump S3a, S3b Off Suma Park Dam Level High Decision Variable
Batch Ponds Level Low Fixed
Stormwater Pump S3a, S3b On Suma Park Dam Level Low Decision Variable
Batch Ponds Level High Fixed
Stormwater Pump S4a, S4b Off Holding Pond Level High Decision Variable
Mitchell Wetland Level Low Fixed
Stormwater Pump S4a, S4b On Holding Pond Level Low Decision Variable
Mitchell Wetland Level High Fixed
Stormwater Pump S5a, S5b Off Holding Pond Level High Decision Variable
Brooklands Wetland Level Low Fixed
Stormwater Pump S5a, S5b On Holding Pond Level Low Decision Variable
Brooklands Wetland Level High Fixed
Stormwater Pump S6a, S6b Off Holding Pond Level High Decision Variable
Somerset Wetland Level Low Fixed
Stormwater Pump S6a, S6b On Holding Pond Level Low Decision Variable
Somerset Wetland Level High Fixed
Groundwater Pump G1 Off Suma Park Dam Level High Decision Variable
Groundwater Pump G1 On Suma Park Dam Level Low Decision Variable
Groundwater Pump G2 Off Suma Park Dam Level High Decision Variable
Groundwater Pump G2 On Suma Park Dam Level Low Decision Variable
Groundwater Pump G3 Off Suma Park Dam Level High Decision Variable
Groundwater Pump G3 On Suma Park Dam Level Low Decision Variable
Energy Optimization Results
Minimization of pump energy use over the longer term is presented here as an example of optimization
of this system. Note that the system is simulated over one year, at a daily time step in EPANET. Additional
computer code was added to the EPANET hydraulic simulation to take into account other process such
as rainfall to and evaporation from storages. This code essentially adds a mass balance component to
the hydraulic simulation. Historical rainfall for the catchments in the system was modelled in MUSIC
hydrologic software to develop inflow series for the ponds and reservoirs. For this optimization the year
with the closest to average rainfall was used, however, other years of rainfall were available and this
optimization could be extended to consider other climate conditions.
NSGAII (Non-dominated Sorting Genetic Algorithm II) software was used for the optimization, with five
random seeds, a population size of 50, 100 generations and probabilities of crossover and mutation of
0.8 and 0.02 respectively. In the best solution found, the system used a total of 793 MWh of energy over
71 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
the entire year. Table 5.10 shows the volume of water pumped from each source to Suma Park Dam (and
supplied from the local catchment) and the energy used by each of the pumps for the optimal solution.
Pumping from the Macquarie is very energy intensive so this is only used at the very end of the simulation
when the level in Suma Park Dam is very low, in order to achieve the end target level constraint (Figure
5.11 and Figure 5.12). Groundwater and stormwater sources are used initially to increase the level of
Suma Park Dam to its maximum, and then not used again until around Day 160 when the level in the dam
has dropped again. Only one of the Macquarie pumps is used, as, despite operating at a lower energy
efficiency point, it uses less energy overall than operating two pumps in parallel. In dryer years, both
pumps may need to be utilized in order to ensure supply to Suma Park Dam. Nearly all of the available
groundwater license is used; G1 and G2 have a combined license of 180 ML /year, and G3 280 ML/year.
Groundwater is more energy intensive than stormwater, however, it can be used at any time throughout
the year, while stormwater is reliant of inflow. Most of the stormwater provided to Suma Park Dam came
from the Blackman’s Creek scheme (S1) rather than the Ploughman’s Creek scheme (S4, S5 and S6).
While the storage capacity of the Blackman’s Creek scheme is much lower, the pump capacity and energy
efficiency is much greater than in the Ploughman’s Creek scheme, so it provides more water.
Table 5.10: Volume of water pumped/supplied and energy used in the optimal energy solution
Energy Rate
Source Pump Volume (ML) Energy (MWh)
(MWh/ML)
M1a 0 0 0
Macquarie River M1b 74 150 2.02
Total 74 150 2.02
G1 24 11 0.46
G2 146 79 0.54
Groundwater*
G3 235 106 0.45
Total 405 196 0.48
S1a 258 39 0.15
S1b 479 71 0.15
S2a 828 65 0.08
S2b 237 21 0.09
S3a 1022 170 0.17
S3b 22 5.5 0.25
Stormwater S4a 178 41 0.23
S4b 12 3.1 0.27
S5a 24 4.8 0.20
S5b 56 11 0.19
S6a 60 11 0.18
S6b 26 5.0 0.19
Total** 1044 447 0.43
Spring Creek and Suma Park Catchment - 3865*** - -
*The energy consumption for the groundwater pumps includes both the transfer and bore pumps, i.e. the energy for Pump G1
includes G1a (not modelling in EPANET, energy use estimated from volume) and G1b (modelled in EPANET)
**The total volume supplied by the stormwater schemes is measured as the combined volume supplied by Pumps S3a and
S3b (which discharge to Suma Park Dam), while the total energy is the total of all pumps.
***This is the volume supplied by the natural catchment for the town’s consumption, the total inflow from the catchment is
greater than this however some is used to provide environmental flows and some spills.
72 |
ADE | Publication 2: Framework for the Optimization of Operation and Design of Systems with Different
Alternative Water Sources
17.5
17.0
16.5
)m
16.0
(
le
v
e
L 15.5
15.0
14.5
14.0
0 16 31 45 56 76 90 124 162 186 202 224 246 289 346 358
Time (Days)
Figure 5.11: Variation in Suma Park Dam level for the energy optimal solution
1200
1000
)
L
M
(
d
800
e
p
m
u 600
P
e
m
u 400
lo
V
200
0
0 16 31 45 56 76 90 124 162 186 202 224 246 289 346 358
Time (Days)
Macquarie River Groundwater Stormwater
Figure 5.12: Volume pumped from each source to Suma Park Dam for the energy optimal solution
5.5 Conclusions
A generalized framework for the optimization of the design and operation of water supply and distribution
systems has been developed and two case study systems have been used as examples of how to apply
it. The framework is comprised of several components; the options component describes the design and
operational decision variables for the optimization, the infrastructure component covers the infrastructure
aspects of the system that need to be modeled and their data requirements, the analysis component
includes the simulation of the system and evaluation against the objectives and constraints, and finally
the government policy component describes the regulations that may affect other aspects of the
framework. These components fit within an optimization algorithm structure, which firstly generates
potential solutions using the decision variables in the options component, models the system according
to the infrastructure component and evaluates potential solutions using the analysis component. The
evaluation of potential solutions then feeds into the solution space which informs how the decision
variables are changed in the next set of potential solutions. Sensitivity analysis of parameters will
significant uncertainty should be undertaken to ensure robust solutions. The framework also applies to
simulation of systems prior to or without optimization.
73 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Abstract
A harvested stormwater and managed aquifer recharge system has been analysed through both
simulation sensitivity analysis and optimization to reduce operational pumping costs and increase the
volume of water harvested. The simulation sensitivity analysis explored increasing the size of a storage
tank, replacing the three harvesting pumps and using wider tank trigger levels in the system operation.
In the optimization, trigger levels and irrigation schedules were considered as decision variables.
Various streamflow (input) series have been considered in the optimization by finding the optimal
controls for each individual series or by finding the controls that best perform under a range of different
conditions. Optimal controls for the current system were compared to those found for the system with
new replacement pumps. The newly sized pumps were found to provide significant benefits by reducing
pump operating costs by 50%, and by increasing the volume of water able to be harvested. Using wider
tank trigger levels and altering the irrigation schedule so that the irrigation pump operated at a more
efficient point also resulted in a small reduction in cost for the current system.
78 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.1 Introduction
As climate change and population growth highlight water security issues, alternative water sources are
increasingly being used to supplement potable supply (Fielding et al. 2015). Harvested stormwater is an
example of such a source, in which runoff from pervious and impervious surfaces (generally in urban
environments) is collected, treated and supplied to consumers (Naylor et al. 2012). Typically,
stormwater is supplied for non-potable end uses such as irrigation of public spaces, household garden
watering or toilet flushing, however, in some cases it has been used for potable supply (McArdle et al.
2011). As well as improving water security, harvested stormwater can have other benefits that
communities place value on such as reduced flooding, improved surface water quality, improved
hydrologic function and improved aquatic habitats (Londoño Cadavid and Ando 2013). Where there is
low understanding of the risks of stormwater to human health, communities may be less likely to accept
harvested stormwater projects and education programs may need to be considered (Hwang et al. 2006).
Water system managers perceive operation and maintenance costs as one of the greatest barriers to
implementation of harvested stormwater projects (Dobbie and Brown 2012). Determining strategies to
reduce ongoing energy costs of these systems is therefore an important task.
Previous studies on the optimization of harvested stormwater systems have usually considered only the
design of the system, not the operation. When harvested stormwater is used to supplement or add to
potable supplies, the yield of the system (volume of water harvested or provided to users) is an
important variable to be maximized (such as in McArdle et al. 2011; Marchi et al. 2016a; di Matteo et al.
2016). McArdle et al. (2011) optimized the design of a harvested stormwater system to minimize life-
cycle costs, maximize yield and minimize the impact of the system on the amenity of a public reserve.
Marchi et al. (2016a) also optimized the design of a harvested stormwater system, which included
Managed Aquifer Recharge (MAR). They included consideration of externalities and climate change,
and found that the values of both the net present value and yield objectives decreased when climate
change impacts were considered.
As well as objectives of minimizing costs and maximizing yield, maximizing water quality is often
included, such as in di Matteo et al. (2016). Studies assessing the performance of harvested stormwater
systems often focus on water quality rather than the cost of energy for pumping (for example, Burns and
Mitchell 2008; Nnadi et al. 2015; Petterson et al. 2016). Labadie et al. (2012) optimized the operation of
a stormwater system, however, the objective was to reduce the environmental impact on the
downstream ecosystem rather than minimization of pumping costs or maximization of the volume of
water harvested.
The remainder of this chapter is organized as follows; firstly, background on a case study system gives
context for the other sections, the methodology of the analysis and optimization of this case study is
then discussed, followed by results of the simulation sensitivity analysis and optimization, and finally
conclusions are drawn.
6.2 Case Study: Ridge Park Managed Aquifer Recharge System
The Ridge Park Managed Aquifer Recharge Scheme in South Australia supplies non-potable water to
sports and recreational areas for irrigation use. South Australia has largely seasonal rainfall, with most
occurring over the winter months around May to October. Water supplies also rely on imported water
from the River Murray, which is costly (due to distance and elevation rise) and highly regulated.
Alternative water source systems are important to reduce use of potable supplies from variable
catchment inflows and the River Murray. The system is located in the metropolitan area of the city of
Adelaide and is operated by the Unley City Council. It was designed to harvest up to 60 ML of
stormwater per year for injection into a confined aquifer, which occurs over the winter, while in summer
79 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
the harvested water is drawn from the aquifer and used for irrigation. Figure 6.1 shows a schematic of
the system, which is described below. Note that the case study analyzed in this research was based on
the best available information for the real-life system. There may be some differences between the
simulated and real-life systems.
Henry Codd
Reserve
Unley Oval
Windsor St
Reserve
Fern Ave
Reserve
Scammel
Reserve
Ferguson Ave
Reserve
Fullarton Park
Treatment
Fraser Plant
Storage Pump 3
Reserve
Tank Ridge Park
Bore
Pump
Pump 2
Pump 1
Dam
Glen Osmond Bioretention
Creek Harvest Pond Basin
Figure 6.1: Schematic of Ridge Park Managed Aquifer Recharge Scheme
In winter, stormwater is harvested from the Glen Osmond Creek, an urban waterway that receives
approximately 200 ML of runoff per year (on average) at the point of harvest. A dam has been
constructed in Ridge Park, in the suburb of Myrtle Bank, to create the 1500 kL Harvest Pond. Water is
then pumped to a Bioretention Basin which provides some treatment, and then pumped to a small
treatment plant that includes a micro-filter and ultra-violet (UV) treatment. Once the water has been
adequately treated, it is stored in a 36 kL above ground tank next to the treatment plant and final pump
station. From the Storage Tank, water is injected into an artesian, fractured rock aquifer for long term
storage.
In summer, when no water is being harvested, water is extracted from the aquifer using the same bore
and stored in the tank, before being pumped or gravity-fed to irrigation points. The Ridge Park Reserve
is irrigated by a pressurized 90 mm diameter irrigation line, as it is at higher elevation than the Storage
Tank. Fraser Reserve is also connected to the pressurized system; although it is at lower elevation, it is
not enough to ensure adequate pressures for irrigation. In total, the pressurized system supplies almost
15 ML per year for irrigation. The remaining seven open space reserves are on a 180 mm diameter
gravity-fed line which supplies a total demand of over 37 ML per year. The total irrigation pipeline length
is 4.3 km. As rainfall and therefore streamflow is variable year to year, the volume harvested will also
vary. On average the harvested volume should be enough to provide the irrigation demand for the
grassed reserves, however, the injection volume is not restricted to the harvest volume from the
previous season. If not enough stormwater was harvested over several winter seasons, potable back-up
supply is available (assuming no water restrictions are in place).
80 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.3 Methodology
The framework presented in Blinco et al. (2017a) has been used to develop the methodology for this
study. Within an optimization algorithm, the framework incorporates the options (decision variables) for
the problem, the water and electricity infrastructure that may need to be modelled, the simulation tools
used to model the system and the analysis of the system in terms of objectives and constraints. Blinco
et al. (2017a) also discuss the importance of sensitivity analysis; as well as finding optimization results
that are robust for different inputs, this process can highlight parameters that are important to, or in
contrast, have little impact on, the results of an optimization problem.
In this research, sensitivity analysis is performed prior to optimization for a range of system
configurations and inputs, by simulating the system in EPANET hydraulic simulation software (Rossman
2000). Performing simulation runs is an extremely important part of the process so that the user can
fully understand the system prior to the investigation of optimization. Results from the simulation
sensitivity analysis are then compared in their absolute form (such as cost or number of pump switches)
and relative to the base case (current operation) as a percentage. These simulation results inform what
is investigated through optimization of the system; the solution space for the optimization may also be
reduced by removing options that had little impact in the simulation sensitivity analysis.
6.3.1 Simulation Model Development
Two models of the case study system have been developed in EPANET; one for the winter operation of
harvesting and confined aquifer injection, and one for the summer operation of confined aquifer
extraction and irrigation. The operation of the bore cannot be switched from injection to extraction
frequently, so the system is operated (and hence modelled) with two distinct seasons. For both models,
assumptions included that minor losses are negligible, the pump and efficiency curves from the
manufacturer catalogue are still valid, and there has been no build-up of biofilm in the pipe systems.
These models did not simulate water quality as the main focus of this study is operational pumping
costs. Both systems were simulated for one week in EPANET, with a 15 minute hydraulic time step. The
simulation time was representative of the full season as multiple streamflow scenarios were considered
in the winter system and the irrigation schedule repeats weekly in the summer. Each year the specific
start and end of each season will vary depending on the weather, however it is assumed that each
season lasts for 26 weeks.
Trigger levels (related to volumes in the three storages as shown in Table 6.1) control when the pumps
in the winter harvesting and injection system (Pumps 1, 2 and 3) turn on and off. During summer, the
Bore Pump is also controlled by trigger levels in the Storage Tank, while Pump 3 is controlled by the
irrigation demands instead of trigger levels. The irrigation schedule is arranged so that different open
space reserves are irrigated on different nights (Table 6.2 and Figure 6.2). Pump 3 is a variable speed
pump (VSP) and is operated at 80% of full speed for injection (such that the flow is less than the 7 L/s
maximum for injection) and 75% of full speed for irrigation (such that the target pressure downstream of
the pump is achieved at the expected demand rates). A peak/off-peak electrical energy price tariff
applied to the entire system; a peak price of 25.53 c/kWh was applied from 7am to 9pm on weekdays,
and an off-peak price of 15.26 c/kWh was applied over night and on weekends. The electricity tariff
pattern assumed the simulation was starting on a Sunday. Blinco et al. (2017a) gives a detailed
description of the development of the simulation models.
81 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Table 6.1: Trigger Levels for the Ridge Park system
Current Setpoint
Storage and Trigger Level Type Start Pump Stop Pump
Volume (%) Level (m)
Harvest Pond High Level 80 1.6 1 -
Harvest Pond Low Level 50 1.0 - 1
Bioretention Basin High Level 90 0.80 2 1
Bioretention Basin Low Level 50 0.59 - 2
Storage Tank High Level 90 2.25 3 2, Bore
Storage Tank Low Level 70 1.75 Bore 3
Note that this table has been taken from Blinco et al. (2017a) and provided here for completeness.
Table 6.2: Irrigation demand schedule for the Ridge Park system
Open Space Reserve Demand Rate (L/s) Duration/day (hr) Start Time Irrigation Days
Ridge Park 1 3.53 8.33 9:30 PM Mon & Wed
Ridge Park 2 3.53 8.67 9:30 PM Tues & Thurs
Fraser Reserve 1.41 5.83 9:30 PM Mon & Wed
Ferguson Ave Reserve 2.00 5.00 9:30 PM Tues & Thurs
Scammell Reserve 2.15 6.00 10:00 PM Tues & Thurs
Fullarton Park 1 3.85 1.67 10:00 PM Mon & Wed
Fullarton Park 2 3.85 6.67 10:00 PM Tues & Thurs
Fern Ave Reserve 3.53 3.33 10:00 PM Mon & Wed
Windsor St Reserve 2.20 8.00 8:30 PM Tues & Thurs
Henry Codd Reserve 1.10 8.00 10:00 PM Mon & Wed
Unley Oval 5.57 9.00 9:00 PM Sun, Mon & Wed
Note that this table has been taken from Blinco et al. (2017a) and provided here for completeness.
20
16
)s 12
/L
(
w
o 8
lF
4
0
Simulation Time (hr)
Total Irrigation Demand Gravity Irrigation Demand Pressure Irrigation Demand and Pump 3 Flow
Figure 6.2: Irrigation schedules under the current operation (note that this figure has been taken from Blinco et al.
(2017a) for comparison to Figure 6.10)
Winter System (Stormwater Harvesting and Confined Aquifer Injection)
The winter simulation model within EPANET included the Harvest Pond, Biorentention Basin, Storage
Tank and Aquifer, and all the pumps and pipes required to transfer water between them (Figure 6.3).
Glen Osmond Creek was included as an input node, with a negative base demand applied to simulate
in EPANET that water should flow into the Harvest Pond. Recorded streamflow data were applied as a
demand pattern to this node. A volume-elevation curve was applied to the Bioretention Basin to account
for the porosity of the filter media and the height of water storage above this. No volume curve
information was available for the Harvest Pond, so it was assumed to have a constant surface area.
82 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Pressure sustaining valves (PSVs) were inserted into the simulation model to take into account the fact
that the discharges to the Bioretention Basin and Storage Tank are from pipes over the top of these
storages. A general purpose valve (GPV) upstream of the Storage Tank took into account the energy
losses through the micro-filter (losses over the UV machine are assumed negligible). A pressure
breaker valve was used to take into account the headloss through the bore during injection. The
minimum head loss over the bore was 3.0 m due to the water being injected around the Bore Pump.
This head loss increased to 4.8 m over the course of 1 week (0.3 m increase per day) as the bore starts
to clog (it was assumed that a backwash of the bore is initiated once a week). The effective water level
of the confined aquifer was estimated to be 5.0 m above the ground surface at the bore pit and the
impressed level during injection another 45.0 m above this.
Figure 6.3: EPANET model of the Winter System (harvesting and confined aquifer injection)
Summer System (Confined Aquifer Extraction and Irrigation)
The summer simulation model within EPANET included the Aquifer and Bore Pump, Storage Tank,
Pump 3 (for irrigation) and the pressure and gravity distribution systems (Figure 6.4). At each open
space reserve, there are small irrigation systems transferring water from the main distribution line to the
sprinkler heads. These pipes were not included in the EPANET model, as the demand information
available was for each open space reserve rather than individual sprinklers, and pressure constraints
were considered just downstream of Pump 3 to ensure there was enough pressure for the sprinklers to
operate effectively. Demands at Ridge Park and Fullarton Park were split into two groups of irrigation
stations so that the irrigation for these areas can be spread out over different nights. As in the winter
model, there was a PSV just upstream of the Storage Tank to account for the inlet being at the top of
the tank. There was also a PSV in the bore headworks which represented an existing valve. The
confined aquifer was modelled as a reservoir, with the head level assumed to be at the effective water
level for extraction. The Bore Pump was not likely to be operated for long enough to create significant
drawdown (Wang et al. 2009).
6.3.2 Optimization Model Formulation
The Non-dominated Sorting Genetic Algorithm II (NSGA-II, Deb et al. 2002) was chosen as it can
incorporate multiple objectives and has been shown to perform well for water distribution system
problems (Wang et al. 2015). NSGA-II was connected with the EPANET Toolkit To Alter Rule-based
controls (ETTAR) developed in Marchi et al. (2016b) to allow the optimization of the operating rules
(trigger levels and irrigation scheduling) for the case study system. ETTAR also incorporates the
variable speed pump (VSP) efficiency correction to allow for accurate calculation of pump energy use
for VSPs (Marchi and Simpson 2013). Simulation sensitivity analysis was performed prior to
optimization, in order to provide a better understanding of the system and refine the optimization
formulation. Objectives and decision variables are introduced here, and further developed after the
results of the sensitivity simulation analysis are presented.
83 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Figure 6.4: EPANET model of the Summer System (confined aquifer extraction and irrigation)
There were two objectives of the optimization problem; firstly to minimize the cost of electricity used to
operate the pumps (Eq. 6.1) and secondly maximize the volume of water harvested over the simulation
period (Eq. 6.2). In many water resources optimization problems, objective functions for operational cost
take into account the volume of water delivered, calculating the cost per unit volume pumped. In this
case, however, the volume harvested is considered as an additional objective function to be maximised,
and therefore does not need to be included in the cost objective function.
𝑂𝐶 = ∑ 𝑇𝑥𝐸 (6.1)
𝑖 𝑖 𝑖
𝑉𝐻 = ∑ 𝑉 (6.2)
𝑖 𝑖
where OC = operational cost (dollars/week); T = electricity tariff for each time step i (dollars/kWh); E =
i i
energy consumption for each time step i (kWh); VH = volume harvested (ML); and V = volume
i
harvested in each time step i (ML). The time step i would range from 1 to 672 for the week long
simulation at 15 minute time increments used for the case study in this research.
Both operational and design decision variables were considered in this paper. Although the case study
system considered in this research had already been constructed, adjustments to the design were
possible, including upgrading the storage tank size and replacing the pumps. Operational decision
variables were in the form of trigger levels in each of the storages that would control the pump
operations. The irrigation schedule was also considered as a decision variable, which required new
computer code to be developed to implement this in NSGA-II. For each open space reserve, two
decision variables and four set variables were defined. The decision variables were the start day for
irrigation (coded as integers with 0 being the starting day for the simulation) and the start time for
irrigation (also integer coded, referring to the time in hours, i.e. 8:30pm would be 20.50 for the
simulation). For each open space reserve, the demand rate (in L/s), duration of irrigation (in hours),
84 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
number of days of irrigation (per week) and the gap between irrigation days (a gap of one day results in
irrigation every second day) were set.
As upgrades to the system infrastructure come at a cost, Net Present Value (NPV) analysis can be used
to determine if operational cost savings achieved with new infrastructure would provide a net financial
benefit. NPV analysis takes into account the costs and returns of a project over time, with future costs
and benefits discounted to current prices, as shown in Eq. (6.3). The operational costs savings realised
by any new infrastructure were treated as returns into the future, and the capital costs of new
infrastructure were assumed to occur at the start of the period and therefore were not discounted. A
positive NPV indicates that a project is financially beneficial, while a negative NPV indicates that it has a
net financial cost.
𝑁𝑃𝑉 = ∑𝑇 𝐶𝑂 −𝐶 (6.3)
𝑡=1(1+𝑟)𝑡 𝑐
where NPV = net present value (dollars); T = time period (years); t = time step (years); C = operating
O
cost returns for one time step (dollars/time step) (in this study the difference in the operating cost with
new infrastructure and the operating cost with current infrastructure); r = discount rate (decimal); and C
c
= capital cost of new infrastructure (dollars).
Two different methods for incorporating different streamflow series were also implemented in the
optimization (Figure 6.5); (1) individual series and (2) looped series. The first considers each streamflow
series individually, which would be most applicable in situations where a good forecast is available and
the operating rules can be easily altered. Optimization of the system is performed with one streamflow
series used in the simulation, if other streamflow series are of interest, the optimization is repeated for
each new series. In this method, if n series are considered, n Pareto fronts will be produced. The
second method loops the streamflow series within the optimization algorithm, generating solutions that
will be robust to many possible realizations. Each potential solution in the optimization is simulated n
times for n streamflow series, however, only one Pareto front is produced. The objective function values
calculated for each of the n simulations of one potential solution are averaged to provide just one value
of each objective function for each solution.
6.4 Simulation Sensitivity Analysis
6.4.1 Simulation Sensitivity Analysis Scenarios
Simulation of the current operation of the system in Blinco et al. (2017a) showed that the pumps were
turning on and off very frequently, which should be avoided to reduce maintenance costs and prevent
general wear and tear of the pumps. One of the problems was that Pumps 1 and 2 are oversized
compared to Pump 3 (the flow into the aquifer is restricted to 7 L/s, however, Pumps 1 and 2 operate at
above 20 L/s). The operation of the system with the current pump curves was compared to that with
newly sized pump curves for Pumps 1 and 2 that will allow them to operate at around 7 L/s. The new
pump curve for Pump 2 was also chosen to significantly improve the efficiency of this pump. Sizing of
Pump 3 was considered; as it was originally designed to supply two bores, the best efficiency occurs
closer to 14 L/s than 7 L/s. A new pump was sized to achieve an operating point that had lower flow (at
full speed) and is closer to the best efficiency point. Sizing of the Bore Pump was not considered, as
while the head range of the current pump was higher than needed for extraction, it is also used to
backwash the bore when injecting, which may have a significantly higher head requirement.
85 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Figure 6.5: Methods for incorporating streamflow series in optimization
Another contributing factor to the frequent pump switches was that the trigger levels controlling the
pumps are close together, and the storages (particularly the Storage Tank) are small. Two sets of
trigger levels were investigated; (1) the current trigger levels (shown in Table 6.1) that only used the top
half of the Harvest Pond and Bioretention Basin, and only 20% of the volume of the Storage Tank, and
(2) wider trigger levels that used 70% of each storage volume (Table 6.3). Increasing the size of the
Storage Tank was also investigated: the current Storage Tank volume of 36 kL was compared to
double, five times and ten times the size of the existing tank.
Table 6.3: Wider trigger levels used in the simulation sensitivity analysis
Setpoint
Storage and Trigger Level Type Start Pump Stop Pump
Volume (%) Level (m)
Harvest Pond High Level 90 1.8 1 -
Harvest Pond Low Level 20 0.4 - 1
Bioretention Basin High Level 90 0.80 2 1
Bioretention Basin Low Level 20 0.26 - 2
Storage Tank High Level 90 2.25 3 2, Bore
Storage Tank Low Level 20 0.50 Bore 3
A total of 20 different simulation sensitivity analysis scenarios (Table 6.4) were considered for the winter
system (harvesting and confined aquifer injection), with different combinations of current or wide trigger
levels, current or new pumps, and Storage Tank sizes. Scenario A used the current values for the
following – trigger levels, pump curves and tank sizes – therefore results from this scenario were
considered to be the baseline for comparing all other scenarios. The summer system (confined aquifer
extraction and irrigation) was simulated with the newly sized Pump 3, a larger tank and wider trigger
levels for the Bore Pump.
Each scenario was simulated six times with six different week-long streamflow series. The streamflow
series selected represented a range of operating conditions (dry, wet or average week and (relatively)
constant or variable flow) (Figure 6.6). Results of the current operation indicated that when the average
flow was above approximately 25 L/s, the injected volume could not be significantly increased. In the
analysis of current operations in Blinco et al. (2017a), it was found that when the average streamflow
was above 25 L/s, there was not a significant increase in the amount of water able to be harvested due
86 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
to restrictions of storage volumes and pump flow rates. Of the six streamflow series selected for
simulation sensitivity analysis, four had average flows lower than 25 L/s to provide a wider range of
results and two had average flows around 25 L/s with different levels of variability. None of the series
used in the simulation sensitivity analysis had average flows significantly greater than 25 L/s.
Table 6.4: Simulation sensitivity analysis scenarios for the Winter System
Trigger Levels2 Pumps3 Storage Tank Size
Scenario
Current Wide Current New 36 kL 72 kL 180 kL 360 kL
HP 50-80% HP: 20-90% Q : 20 L/s Q : 7 L/s Current
1 1
BB 50-90% BB: 20-90% Q : 25 L/s Q : 7 L/s
2 2
ST 70-90% ST: 20-90% Q : 7 L/s Q : 7 L/s
3 3
A X X X
winter
B X X X
winter
C X X X
winter
D winter1 X X X
E X X X
winter
F X X X
winter
G X X X
winter
H X X X
winter
J X X X
winter
K X X X
winter
L X X X
winter
M X X X
winter
N X X X
winter
P X X X
winter
Q X X X
winter
R X X X
winter
S
winter
X 34 1, 2 X
T X 3 1, 2 X
winter
U X 3 1, 2 X
winter
V X 3 1, 2 X
winter
1Boxed items correspond to scenarios with the ‘best values’ in Table 6.5.
2Trigger levels for the Harvest Pond (HP), Bioretention Basin (BB) and Storage Tank (ST).
3Typical pump operating flow rates for the current and new pump models.
4In Scenarios S
winter
– V winter, new pump models were considered for Pumps 1 and 2 only, with the current model used for
Pump 3.
70.0
)s 60.0
/L
(
w
50.0
o
lfm
Series 4
a 40.0
e
r
t
S
e
30.0 Series 6
lb
a
lia 20.0 Series 2 Series 5
v
A Series 1
10.0
Series 3
0.0
1 2 3 4 5 6 7
Day
Series 1 Series 2 Series 3 Series 4 Series 5 Series 6
Figure 6.6: Streamflow series used in the simulation sensitivity analysis
87 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.4.2 Simulation Sensitivity Analysis Results – Winter System (Stormwater Harvesting and
Confined Aquifer Injection)
For each simulation, the cost of pumping, volume pumped into the confined aquifer and number of
pump switches were calculated. The results for each scenario were then averaged over the different
streamflow series (Table 6.5 and Table 6.6) and the results for each streamflow series were averaged
over the different scenarios (Table 6.7). Where the volume harvested is below 2.0 ML per week and
therefore below 52 ML (the total irrigation volume) in the season (26 weeks), additional water from the
aquifer may be drawn over the summer period depending on the extraction and injection levels in the
previous year. Potable back-up supply can also be used if required.
Simulation Results for Changes in Trigger Levels, Pump and Storage Volumes
Table 6.5 shows the cost rate (in c/kL) for each pump and overall, the total cost over a week of
operation, the total volume of stormwater injected to the confined aquifer over a week of operation, and
the number of pump switches per day for each. The highlighted cells show the ‘best’ value for each
result variable (for most of the variables this is the minimum, however, for the volume injected it is the
maximum). In all scenarios, the operation of the new pumps was less expensive than the current
pumps, with cost rates around 4-5 c/kL of water injected compared to 8-9 c/kL of water injected. The
overall and individual pump cost rates were lowest in Scenario M , which used the wider trigger
winter
levels and the second largest tank size as well as the new pumps. Scenario M also had the best
winter
cost rate overall and for Pump 3. In terms of total cost per week, Scenario D was the least
winter
expensive, however, this was partly due to a reduced volume of stormwater injected. Incorporating all
possible changes to the system – the wider trigger levels, the new pumps and the largest Storage Tank
size in Scenario R gave the best results in terms of the volume of stormwater injected and number
winter
of pump switches. There were slight differences in the scenarios that resulted in the best values across
the streamflow series, however, the overall trends were very similar.
88 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Table 6.5: Comparison of simulation results for changes in trigger levels, pump sizes and storage volumes
What has been changed? Cost (c/kL) Pump
Cost Volume Switches
Scenario Trigger Storage Pump Pump Pump
Pumps Total3 ($/wk) (ML/wk) (/day)
Levels Tank 1 2 3
Pump 1,2,3
A
winter
36 kL2 0.677 2.244 5.780 8.849 219 2.50 74,79,3
B Wide 36 kL 0.663 2.212 5.921 8.804 209 2.42 70,69,1
winter
C New 36 kL 0.390 0.453 3.643 4.503 143 3.16 54,49,4
winter
D winter1 Wide New 36 kL 0.381 0.447 3.568 4.408 138 3.11 47,43,1
E 72 kL 0.665 2.252 5.795 8.717 216 2.49 78,74,2
winter
F Wide 72 kL 0.657 2.273 5.916 8.853 212 2.44 72,67,0
winter
G New 72 kL 0.389 0.454 3.633 4.492 143 3.17 53,47,2
winter
H Wide New 72 kL 0.379 0.445 3.528 4.360 141 3.26 46,42,1
winter
J 180 kL 0.665 2.278 5.810 8.693 215 2.49 75,67,1
winter
K Wide 180 kL 0.656 2.251 5.797 8.736 214 2.48 70,63,0
winter
L New 180 kL 0.390 0.453 3.630 4.490 144 3.20 48,46,1
winter
M Wide New 180 kL 0.368 0.426 3.379 4.187 145 3.38 40,38,0
winter
N 360 kL 0.661 2.289 5.061 8.774 215 2.48 75,68,1
winter
P Wide 360 kL 0.658 2.273 5.622 8.603 217 2.54 67,61,0
winter
Q New 360 kL 0.391 0.459 3.662 4.537 143 3.17 49,44,1
winter
R Wide New 360 kL 0.374 0.432 3.421 4.233 152 3.51 34,32,0
winter
S
winter
1, 24 36 kL 0.392 0.455 5.810 6.674 163 2.46 104,109,3
T Wide 1, 2 36 kL 0.387 0.451 5.857 6.702 159 2.41 99,105,1
winter
U 1, 2 72 kL 0.393 0.455 5.810 6.670 163 2.46 101,109,2
winter
V Wide 1, 2 72 kL 0.385 0.445 5.707 6.546 159 2.44 93,100,0
winter
1Boxed cells represent the ‘best values’ for each variable, scenarios that resulted in these ‘best values’ are boxed here and in
Table 6.4 and Table 6.6.
2Current Storage Tank size is 36 kL.
3The total cost rate is calculated as the average of the individual cost rates for each streamflow series, rather than the
average cost per week divided by the average volume per week.
4In Scenarios S
winter
– V winter, new pump models were considered for Pumps 1 and 2 only, with the current model used for
Pump 3.
Comparison of Simulation Results to Scenario A as Baseline Case
winter
Using Scenario A as a baseline (Table 6.6) shows that replacing the pumps has the most significant
winter
impact on cost, while the other changes result in only minor cost reductions. The new pumps also have
the most significant impact on reducing the number of pump switches, however, using wider trigger
levels and increasing the Storage Tank size (to five or ten times the current size) does also have some
effect. Doubling the Storage Tank size does not have a significant impact on either cost or pump
switches. The percentages of volume pumped and cost of energy in the peak and off-peak times do not
vary significantly for the different scenarios. Slightly less volume is pumped in the peak time (there are
70 peak hours in the week and 98 off-peak hours), with peak volumes ranging from 43-49% of total
volume and off-peak volumes ranging from 51-57%. The cost of pumping in the peak time is greater
than that in off-peak, 57-61% of total cost occurs in peak times compared to 39-43% in off-peak,
because of the higher electricity price.
89 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
Table 6.6: Comparison of simulation results for Scenarios B -V to that of Scenario A (Baseline Case)
winter winter winter
What has been changed?
Total Cost Pump Switch Cost to Harvest Diff. in Cost to
Scenario Trigger Storage
Pumps Diff. (%) Diff. (%) 3 ML ($) Harvest 3 ML (%)
Levels Tank
A
winter
236 kL 0 0 265 0
B Wide 36 kL -4 -10 264 -1
winter
C New 36 kL -35 -31 135 -49
winter
1D
winter
Wide New 36 kL -37 -41 132 -50
E 72 kL -1 -1 262 0
winter
F Wide 72 kL -3 -11 266 -1
winter
G New 72 kL -35 -35 135 -49
winter
H Wide New 72 kL -36 -43 131 -50
winter
J 180 kL -2 -8 261 -2
winter
K Wide 180 kL -2 -15 262 -1
winter
L New 180 kL -34 -39 135 -49
winter
M Wide New 180 kL -34 -50 126 -53
winter
N 360 kL -2 -8 263 -1
winter
P Wide 360 kL -1 -17 258 -3
winter
Q New 360 kL -35 -40 136 -49
winter
R Wide New 360 kL -30 -58 127 -52
winter
S
winter
31, 2 36 kL -25 +38 200 -25
T Wide 1, 2 36 kL -27 +46 201 -24
winter
U 1, 2 72 kL -25 +97 200 -25
winter
V Wide 1, 2 72 kL -27 +341 196 -26
winter
1Boxed items correspond to scenarios with the ‘best values’ in Table 6.5.
2Current Storage Tank size is 36 kL.
3In Scenarios S
winter
– V winter, new pump models were considered for Pumps 1 and 2 only, with the current model used for
Pump 3.
Comparison of Simulation Results for Different Streamflow Series
Table 6.7 compares the results averaged over all scenarios for each streamflow series. A higher
average flow in a streamflow series does not necessarily mean that the volume of water injected will be
greater; the variability of the flow and the number of days with a flow rate less than 7 L/s (the maximum
confined aquifer injection rate) also has an impact. Series 1 and Series 6 both have flows consistently
above 7 L/s; a large increase (157%) in the average flow rate from Series 1 to Series 6 results in a small
increase (8%) in the volume harvested. Series 4 and Series 6 have similar average flow rates, however,
Series 4 has two days with flow rates of less than 7 L/s, which results in a 19% reduction in the volume
of stormwater harvested. Series 4 and Series 5 both have two days with flows below 7 L/s, the average
flow rate for Series 4 is almost double (93% increase) that of Series 5, however, the volume of
stormwater harvested for Series 4 is only slightly less (6%) than that for Series 5. This is caused by the
variability of flow in Series 4, which has a standard deviation 153% times than that of Series 5. As
expected, the total cost of pumping for each series increases with the volume of water harvested and
injected.
Table 6.7: Comparison of simulation results for each streamflow series (averaged across all scenarios A -V )
winter winter
Average Number of Days
Streamflow Standard Cost Rate Total Cost Volume Injected
Streamflow with Flow < 7
Series Deviation of Flows (c/kL) ($/wk) (ML/wk)
(L/s) L/s
1 9.90 0.81 0 6.287 208 3.30
2 10.2 8.09 4 6.406 117 2.77
3 2.79 1.44 7 6.631 72 1.09
4 26.4 23.5 2 6.489 187 2.88
5 13.7 9.30 2 6.308 193 3.05
6 25.4 2.88 0 6.087 217 3.56
90 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.4.3 Simulation Sensitivity Analysis Results – Summer System (Confined Aquifer Extraction
and Irrigation)
Results from the simulation sensitivity analysis of the winter system suggested that increasing the tank
size would not provide a significant pumping cost reduction. Moreover, space restrictions of the site
mean that it is unlikely that increasing the tank size by five or ten times would be considered worthwhile
and it is also likely to be very expensive. Therefore, in the simulation sensitivity analysis of the summer
system, only the current and doubled tank sizes have been considered. This resulted in eight scenarios
with different combinations of current or wide trigger levels in the Storage Tank, current or new Pump 3,
and a Storage Tank size of 36 kL or 72 kL (Table 6.8).
Table 6.8: Simulation sensitivity analysis scenarios for the summer system
Trigger Levels Pumps Storage Tank
Scenario
Current Wide Current New 36 kL 72 kL
ST: 70-90% ST: 20-90% Q
3
= 7 L/s 3Q
3
= 7 L/s
A X X X
summer
1B
summer
X X X
C X X X
summer
1D
summer
X X X
E X X X
summer
2F
summer
X X X
G X X X
summer
H X X X
summer
1For scenarios B
summer
and D summer, a lower trigger level of 40% was used because with a lower trigger level of 20%, the tank
will drain when the demands are greater than the bore pump flow.
2Boxed items correspond to scenarios with the ‘best values’ in Table 6.9.
3Pump operating at a higher efficiency point.
There was minimal difference in the results for most variables except for the number of switches for the
Bore Pump (Table 6.9). As the irrigation demands remain the same, so does the operation of Pump 3
(although there is a slight difference in cost between the current and new Pump 3) and the volume of
water that needs to extracted by the Bore Pump. When the Storage Tank size increased or the trigger
levels were widened, the number of switches required by the Bore Pump was reduced. As all the
irrigation occurred overnight, the times when the Storage Tank required filling are in blocks and so the
operation of the Bore Pump was directly related to the operating capacity of the tank.
Table 6.9: Simulation sensitivity analysis results for the summer system
What has been changed? Cost (c/kL) Volume Pump Switches
Cost
Scenario Trigger Storage Bore Pump Extracted (/day) Bore
Pumps Total ($/wk)
Levels Tank Pump 3 (ML/wk) Pump, Pump 3
A
summer
236 kL 3.516 3.966 4.682 90.31 1.93 16,1
B Wide 36 kL 3.509 3.966 4.688 89.45 1.91 7,1
summer
C New 36 kL 3.516 3.892 4.663 89.70 1.92 16,1
summer
D Wide New 36 kL 3.509 3.892 4.660 89.40 1.92 7,1
summer
E 72 kL 3.509 3.966 4.679 90.00 1.92 8,1
summer
1F
summer
Wide 72 kL 3.509 3.966 4.695 89.09 1.90 2,1
G New 72 kL 3.509 3.892 4.657 89.58 1.92 8,1
summer
H Wide New 72 kL 3.509 3.892 4.672 88.67 1.90 2,1
summer
1Boxed cells in represent the ‘best values’ for each variable, scenarios that resulted in these ‘best values’ are boxed here and
in Table 6.8.
2Current Storage Tank size is 36 kL.
91 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.5 Optimization
6.5.1 Revised Optimization Model Formulation
The formulation of the optimization problem was revised on the basis of the sensitivity analysis results.
They showed that there was little benefit in increasing the size of the storage tank, therefore
optimization of pump operations was considered with only the current tank size either with (A) the
current pumps or (B) the newly sized pumps. In the winter system, both the cost and volume objectives
were considered. The cost was calculated as the cost of energy used by Pumps 1, 2 and 3 to the
confined aquifer divided by the volume pumped by Pump 3 (in units of c/kL) and the volume objective
was measured as the volume pumped by Pump 3. In this case, the cost objective was calculated
relative to the volume pumped so that it was more easily comparable to the cost of potable water.
For the summer system, only the cost objective was considered, and it was calculated as the total cost
of energy used by the Bore Pump and Pump 3. As the system pumps to meet demand, the volume
pumped does not change between different solutions and therefore it was not necessary to take it into
account in the objective functions. Different potential solutions may have resulted in different storage
tank levels at the end of the summer irrigation period, however, it was considered undesirable to have
more water in the Storage Tank at the end of summer than at the start, as this water would then be
pumped back into the confined aquifer when the winter harvesting season started. As extraction from
and injection to the confined aquifer are both energy intensive, solutions that extracted more water than
was required in summer were not as good as those that extracted the exact demand amount.
Constraints were applied for a maximum number of pump switches of 48 per day (less than the current
operation) for all pumps, a maximum pressure of 45 m and minimum velocity of 1.1 m/s (equivalent to
flow of 7 L/s) downstream of Pump 3 when injecting and a maximum pressure of 40 m downstream of
Pump 3 when irrigating.
For the winter system, there were six trigger level decision variables to be optimized (Table 6.10) and
four trigger level values that were set and not optimized. Possible choices for the trigger level values
ranged from 10% to 100% of the storage volumes, in 10% increments. For the summer system, there
were two trigger level decision variables and 22 irrigation scheduling decision variables (two each for 11
open space reserves). In fact, given that the demand rate and duration for each open space reserve
(Table 6.2) were set, and the number of days per week, only the start day and time of the irrigation need
to be found by the optimization process. Note that the number of days between each irrigation event
was fixed for all open space reserves in the system and set equal to one (i.e. irrigation occurs every
second day). All open space reserves excluding Unley Oval were irrigated twice a week, and had
choices of initial irrigation days of Monday or Tuesday. Unley Oval was irrigated three times a week and
could only start irrigation on Sunday. Possible start times for all open space reserves ranged from 8pm
to 11:30pm in 30 minute increments. For the summer period, the Bore Pump was controlled by the two
trigger level decision variables, which were levels in the Storage Tank (ranging from 10% to 100% in
10% increments).
Table 6.10: Optimization decision variables and set trigger levels for the winter system
Condition(s)
Rule Effect Restriction
Set Optimized
1 Harvest Pond level > 0.2 AND Bioretetion level < a1 Pump 1 On
b > a
2 Harvest Pond level ≤ 0.2 OR Bioretention level ≥ b Pump 1 Off
3 Bioretetion level > 0.13 AND Storage Tank level < c Pump 2 On
d > c
4 Bioretention level ≤ 0.13 OR Storage Tank level ≥ d Pump 2 Off
5 - Storage Tank level < e Pump 3 Off
f > e
6 - Storage Tank level ≥ f Pump 3 On
1Decision variables are a, b, c, d, e and f.
92 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
6.5.2 Optimization Results – Winter System (Stormwater Harvesting and Confined Aquifer
Injection)
Using the looped streamflow method (with the streamflow series in Figure 6.6), Pareto fronts were
developed for both system configurations (current and new pumps) as shown in Figure 6.7. Note that in
all of the Pareto fronts presented in this section, the ‘best’ solution would be the one closest to the top
left corner of the plot (maximizing volume harvested on the vertical axis and minimizing the cost rate on
the horizontal axis). Moving from the front for System A to that for System B provides a large
improvement in the Pareto solutions, which indicates the effect of replacing the pumps. The new pumps
were also able to harvest more water, with the front for System B extending to over 3.5 ML/week, while
the fronts for System A did not quite reach 3.0 ML/week. In order to supply all of the summer irrigation
demands from harvested stormwater (therefore not using potable supply), a harvest volume of 2.0
ML/week is required on average.
4.0
)k
e-
s
w
3.5
eo
w /Llfm 3.0
Ma
(
d
e
tse r
ts
r
e
22 .. 05
ev
vo
r
a d 1.5
He
e
mg
a r 1.0
u lo Ve v a
0.5
0.0
0 1 2 3 4 5 6 7 8 9 10
Cost Rate (c/kL) -averaged over streamflows
A: Current System B: New Pumps System A Solution for Comparison* System B Solution for Comparison*
Figure 6.7: Pareto fronts for both system configurations using the looped streamflow method
*Solutions compared to current operation in Table 6.11
The Pareto fronts produced from the individual streamflow method showed small differences between
the streamflow series for System A (Figure 6.8) and almost no difference for System B (Figure 6.9).
Streamflow Series 3 showed the largest difference in the Pareto optimal solutions compared with the
other series. This series had flows consistently below 7 L/s (the maximum confined aquifer injection
rate), and therefore the system could not harvest as much when this series was used. For all of the
other streamflow series, the average inflow was above 7 L/s, and while the variability of flow and
number of days with flow below 7 L/s made a difference in the simulation sensitivity analysis, little
impact is shown in the optimization results. The individual streamflow method may show more variability
in results for systems that have a capacity much higher than the average streamflow.
For each system configuration using the looped streamflow method, a solution from the Pareto front that
represented a good trade-off between the objectives was chosen for comparison to the current
operation (Table 6.11, note that the selected solutions are highlighted in Figure 6.7). System A shows a
small improvement, while the new pumps in System B shows a significant cost rate reduction of 50%.
93 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
3.5
)k
e 3.0
e
w
/L
2.5
M
(
d
e 2.0
ts
e
v
r a 1.5
H
e
m 1.0
u
lo
V
0.5
0.0
0 1 2 3 4 5 6 7 8 9 10
Cost Rate (c/kL)
Series 1 Series 2 Series 3 Series 4 Series 5 Series 6
Figure 6.8: Pareto fronts for System A using the individual streamflow method
4.5
)k 4.0
e
e
w 3.5
/L
M
3.0
(
d
e
ts 2.5
e
v
r 2.0
a
H
e 1.5
m
u
lo 1.0
V
0.5
0.0
0 1 2 3 4 5 6 7 8 9 10
Cost Rate (c/kL)
Series 1 Series 2 Series 3 Series 4 Series 5 Series 6
Figure 6.9: Pareto fronts for System B using the individual streamflow method
Table 6.11: Comparison of cost and volume harvested of Pareto optimal solutions to current operation for the winter
system
Current System A Optimal Solution System B Optimal Solution
Variable
Operation1 Actual Difference Actual Difference
Cost Rate (c/kL) 8.85 8.71 -1% 4.34 -51%
Volume Harvested (ML/week) 2.50 1.84 -26% 3.63 +47%
1Data for the current operation is taken from the simulation sensitivity analysis Scenario A
6.5.3 Optimization Results – Summer System (Confined Aquifer Extraction and Irrigation)
Cost reductions for the summer system could be achieved both with the current pumps and by replacing
Pump 3 (Table 6.12). The optimal solutions for both systems use trigger levels of 0.25 m (10%) and
2.25 m (90%) in the Storage Tank to control the bore pump. These are much wider than the current
trigger levels, utilizing 80% of the Storage Tank volume rather than 20%. Optimal irrigation schedules
for both systems have the two Ridge Park stations and Fraser Reserve (i.e. all demand points on the
pressure line) irrigated on the same night. For both the current Pump 3 and the new Pump 3,
efficiencies are improved with the higher flow rate of all three of the pressure demand points rather than
the flow rate required for only one or two demand points. Irrigation of some open space reserves on the
gravity line was then deferred to other nights, in order to distribute the irrigation more evenly, preventing
94 |
ADE | Publication 3: Optimization of pumping costs and harvested volume for a stormwater harvesting system
the Storage Tank from draining if the Bore Pump could not keep up with higher demands (Figure 6.10
compared to Figure 6.2).
Table 6.12: Comparison of cost of Pareto optimal solutions to current operation for the summer system
System A Optimal Solution System B Optimal Solution
Variable Current Operation
Actual Difference Actual Difference
Cost ($/week) $90.3 $85.3 $82.4
-6% -9%
Cost Rate (c/kL)1 4.74 4.47 4.32
1The cost rate is based on the volume irrigated, which is the same for all solutions
20
16
)s
/L
12
(
d
n
a m 8
e
D
4
0
Simulation Time (hr)
Total Irrigation Demand Gravity Irrigation Demand Pressure Irrigation Demand and Pump 3 Flow
Figure 6.10: Irrigation schedules for the optimized operation for System B
While installing new pumps would be a significant investment, the amount of operational savings may
make it worthwhile to the system manager. The difference in cost rate from the optimized operation with
the current pump to the optimized operation with new pumps is 4.43 c/kL in winter and 0.15 c/kL in
summer. If the full irrigation demand of 52 ML is harvested in winter and supplied in summer, this
amounts to $2381 in savings per year. The cost of the newly sized pumps was estimated to be just
under $9000. Using a discount rate of 6% over a 20 year period, the net present value of replacing the
pumps comes to over $18 000. This indicates that replacing the pumps would be financially beneficial
for the Council.
6.6 Conclusions
The operation and system configuration of a harvested stormwater and managed aquifer recharge
system has been thoroughly analyzed both through simulation sensitivity analysis and optimization.
Simulation of the system was split between the winter operation of harvesting and confined aquifer
injection and summer operation of confined aquifer extraction and irrigation. The simulation sensitivity
analysis considered replacing the pumps with smaller, more efficient models, increasing the size of the
Storage Tank and using wider trigger levels. Replacement of the pumps with smaller models was also
considered in the optimization of the system. Different streamflow (input) series have been investigated
using two methods in the optimization; firstly by performing separate optimizations (and therefore
developing separate Pareto fronts) for each series, and secondly by looping the streamflow series within
the optimization to find robust solutions.
Simulation sensitivity analysis of the system found that increasing the size of the Storage Tank would
not provide significant benefits, however, installing smaller pumps with better efficiencies could reduce
costs by 30-37%. Optimization with these new smaller pumps could provide further savings of up to
95 |
ADE | Chapter 7 Conclusions and Future Work
7.1 Thesis Summary
Water supply and distribution systems are a critical part of our society. As climate change and a growing
population put pressure on existing supplies, alternative sources such as harvested stormwater are
becoming more commonly used. Energy use of pumps is a significant concern for water supply systems,
both in terms of cost of electricity and emissions of GHGs. Pump operations have been extensively
analysed and optimized for traditional water distribution systems, however, complex pump operating rules
have not previously been considered. Optimization techniques have not previously been applied to the
minimization of cost of pump operations in alternative water source systems. These systems are generally
more complex to simulate and optimize, as there are additional processes, such as streamflow, and
additional components, such as treatment wetlands that may need to be considered. This thesis
addressed these gaps through the following six objectives developed in Chapter 1:
Objective 1. To develop a framework to optimize alternative water system pump operations for multiple
objectives including minimizing cost and maximizing volume harvested.
Objective 2. To apply the use of new rule-based controls in a modified EPANET2 programmer’s toolkit
to optimize complex pump operational strategies using a combination of trigger levels and
scheduling, and variable trigger levels.
Objective 3. To optimize pumping operations and irrigation scheduling for short time horizons for
systems using harvested stormwater with aquifer storage and recovery and multiple
pumping stations.
Objective 4. To demonstrate the importance of performing detailed simulation analysis of water systems
in order to better understand the system and to inform optimization of the system.
Objective 5. To analyse the sensitivity of optimal pump operations to changes in streamflow (system
inflow) and system design in a stormwater harvesting system.
Objective 6. To minimize GHG emissions from pump operations where operational characteristics are
considered as decision variables and characterize trade-offs between optimal cost and
optimal GHG solutions for these problems.
Optimization of five different types of pump operating controls has been performed on two potable water
distribution system case studies using rule-based controls in EPANET. Minimization of energy costs and
GHG emissions were considered separately using a single-objective genetic algorithm. VSP scheduling
was found to perform better than the other types of pump operating controls, and significant cost savings
were achieved for the real-life South Australian case study. A framework has been developed to
demonstrate how these types of optimization tools could be applied to water systems that use alternative
sources. The framework incorporates design and operational options, water and electrical energy
infrastructure, simulation models, government policy, and objectives and constraints within an
optimization algorithm process. This framework was then applied to pump operations in a integrated
supply system with multiple alternative water sources and a harvested stormwater system, in order to
minimize pump energy costs and maximize the volume of water harvested. An extensive simulation
sensitivity analysis was performed on a case study stormwater system, demonstrating the importance of
pump selection. Optimization of the system found optimal pump operating strategies for both individual
streamflow (input) series and solutions that were robust to multiple streamflow series. As well as replacing
the pumps in the system, altering the tank trigger levels and irrigation schedule provided a reduction in
pump operating costs.
99 |
ADE | Conclusions and Future Work
7.2 Research Contributions
The overall contribution of this research is the application of pump operations optimization techniques that
have been successfully developed on traditional potable WDSs to systems that utilize alternative water
sources such as harvested stormwater. From the publications presented in Chapters 4 to 6 of this thesis,
the following key contributions have been made to address the knowledge gaps identified in Section 2.4:
The first contribution is the development of a framework for the optimization of systems using alternative
water sources (Objective 1). This framework describes a methodology for optimization of both design
and operations of water systems that use alternative water sources, incorporating options (decision
variables), infrastructure, simulation models, and analysis of objectives and constraints. It also identifies
interactions between different system components, in particular the integrated nature of water and energy
systems, as well as the influence of government policies. Other frameworks and methodologies presented
previously have not covered the same extent of both supply and distribution sides of WDSs, or the range
of alternative water sources considered in this framework. The framework is generalized, and while its
application to two case studies for optimal pump operations is demonstrated in Chapter 5, it could be
used for both design and operations of other alternative water source systems.
The second contribution is an improved understanding of the optimization of complex pump operating
rules including the combination of trigger levels and scheduling (Objective 2). Application of the new
EPANET Toolkit To Alter Rule-Based Controls (ETTAR) allowed five different pump operating control
cases to be investigated for two case study systems in Chapter 4. Previous studies have considered
trigger levels and scheduling separately, or where combined trigger levels and scheduling were used,
only one was formulated as a decision variable, with the other being set before optimization.
Another contribution from Chapter 4 is the minimization of GHG emissions for pump operations
(Objective 6). GHGs have been extensively investigated in WDSs, however, this has mainly been in the
optimization of the design of systems. These studies do consider pump operations in order to determine
life-cycle GHG emissions, however, the operating rules are not considered as decision variables. The
work in Chapter 4 minimizes GHG emissions for existing systems, where pump operating rules are
considered as decision variables, rather than the design of the system.
The fourth contribution is the application of optimization techniques developed on traditional WDSs to the
operation of a harvested stormwater system (Objective 3). Optimization of systems using alternative
water sources has not been as extensive as for traditional WDSs, and minimization of pumping costs for
systems harvesting stormwater for re-use has not been previously considered. Chapter 6 extends the
work done on potable WDSs to a harvested stormwater system, which is more complex to simulate and
therefore to optimize. Irrigation scheduling was optimized along with the pump operating rules; in
traditional WDSs, consumer demands cannot be controlled or perfectly predicted and as such represent
a constraint or uncertain variable for the system. In systems that use alternative sources for non-potable
uses such as irrigation of public spaces, the demands may be controlled by the system managers and
therefore considered as decision variables.
The final contribution of this thesis is the demonstration of the importance of extensive pre-optimization
simulation and analysis of water systems (Objective 4). Before optimization was performed on the case
study system in Chapter 6, extensive simulation sensitivity analysis was used to refine the optimization
problem. Sensitivity of the system to changes in the pump selection, tank size, trigger levels and
streamflow was rigorously tested (Objective 5). This helped to fully understand the system and to refine
the search space of the optimization.
100 |
ADE | Conclusions and Future Work
7.3 Recommendations for Future Research
Alternative water source systems are more complex to simulate than traditional WDSs and have different
modelling requirements. The harvested stormwater case study investigated in this thesis was simulated
in EPANET hydraulic simulation software, however, this may need to be connected to hydrological or
hydrogeological models for other systems. For the case study presented in Chapter 6, streamflow data
was available and this could easily be implemented as an input to the hydraulic model. Generally,
streamflow data is much more limited than rainfall data, and therefore other systems may only have rainfall
data available. In this case, a hydrological rainfall-runoff model would need to be used to provide input to
the hydraulic model. Hydrologic models could also be utilized in order to consider the impacts of climate
change on the stormwater runoff volumes and harvesting capacity of stormwater catchments.
Assumptions made about the groundwater aquifer in the case study also meant that it could be
represented purely through the hydraulic model, however, in systems with more complex ground and
surface water interactions, a hydrogeological model may be required. In order to make the methodology
used in the research more generally applicable to other alternative water source systems, hydrological
and hydrogeological simulation should be incorporated.
Water quality is another important consideration, for both traditional potable supply and alternative water
sources that could be included in the future. This may be done through hydraulic simulation; many
programs have at least the ability to calculate water age, if not chemical concentrations as well. Additional
code added on to hydraulic simulation or other programs already available for water quality modelling
could also be required to accurately account for water quality.
As the focus for this research was on the pumping operation of existing systems in the current climate
conditions, there was limited investigation of different streamflow or demand scenarios. Both of these
factors are uncertain now and into the future, particularly when climate change is considered. The
methodology used for the harvested stormwater case study allowed multiple streamflow inputs to be
considered, however, only a small number of recorded data series were used. To make the optimal
solutions more robust to current and future variation in streamflow and demand, multiple replicates based
on statistics of recorded data and projections should be incorporated. This could be achieved by
connecting the methodology in this research with algorithms such as Monte Carlo simulation.
Electricity tariffs are also uncertain into the future; while specific case studies have given electricity tariff
structures and prices for the short-term, energy infrastructure and markets will change in the future
resulting in different electricity tariffs. The case studies in this research all had relatively simple peak and
off-peak electricity tariffs, and one also considered peak demand charge. More complex tariffs such as
those with shoulder periods or different weekend tariffs would increase the complexity of the optimisation
and should be considered in the future.
The framework presented in Chapter 5 discusses many different types of alternative water sources,
however, only harvested stormwater was investigated further in this thesis. A natural extension of this
work would be to apply the framework and methodology to other types of alternative water source
systems, such as recycled wastewater, groundwater and imported water. These different sources will
each have unique components that need to be simulated, which would not be incorporated in the current
methodology developed for the harvested stormwater system. Applying the framework to systems with
more complex pumping arrangements would also help to further develop the methodology.
Further development of the methodology on different types of alternative water source systems would
make it more generalized and allow easier application to all types of alternative water source systems in
the future. An explicit mathematical model of the framework could be developed to allow other researchers
101 |
ADE | Abstract
The understanding and effective management of flood and drought issues within
catchments, are critical to sustaining such systems and the environments they
support. Surface water and groundwater systems within catchments exhibit
important feedbacks and therefore must be considered as a single resource.
Holistic consideration of these systems in catchment hydrology requires the
understanding and quantification of both surface and subsurface flow processes
and their interactions. This requires that the physics driving the
interactions/processes are well understood. Consequently, a need has arisen for
physics-based models that can aid in building intuition about these
interactions/processes, and also assist in quantifying these interactions/processes.
In the last decade, physics-based fully integrated surface-subsurface flow models
have become an important tool in understanding and quantifying flow generation
processes and surface-subsurface interactions. However, due to the relatively short
history of fully integrated models, the analysis and interpretation of outputs is
often incommensurate with the spatiotemporal information within the outputs. A
key shortcoming of these models is the inability to use model outputs to properly
analyse and interpret flow generation mechanisms and surface water-groundwater
interactions with respect to the streamflow hydrograph.
In this research, a new Hydraulic Mixing-Cell (HMC) method for quantifying in-
stream and overland flow generation mechanisms within physics-based models of
surface-subsurface flow is developed. The HMC method is implemented and
tested within the fully integrated surface-subsurface flow model code
HydroGeoSphere. The HMC method is used in a series of applications to quantify
the contributions to total streamflow of groundwater discharge to the stream and
hillslope, and direct rainfall to the stream and hillslope.
Application of the HMC method to a hypothetical catchment is used to investigate
the importance of in-stream flow travel time and losses. Results showed that it is
necessary to account for in-stream travel time and stream losses in order to
accurately quantify the contribution of groundwater to streamflow. The HMC
method is then used with another hypothetical catchment model to investigate the
potential error in 10 commonly used automated baseflow separation methods.
vii |
ADE | Statement of Originality
This work contains no material which has been accepted for the award of any
other degree or diploma in any university or other tertiary institution to Daniel
Partington and, to the best of my knowledge and belief, contains no material
previously published or written by another person, except where due reference has
been made in the text.
I give consent to this copy of my thesis when deposited in the University Library,
being made available for loan and photocopying, subject to the provisions of the
Copyright Act 1968.
The author acknowledges that copyright of published works contained within this
thesis (as listed below) resides with the copyright holders of those works.
List of works:
Partington, D., P. Brunner, C. T. Simmons, R. Therrien, A. D. Werner, G. C.
Dandy, and H. R. Maier. 2011. A hydraulic mixing-cell method to quantify the
groundwater component of streamflow within spatially distributed fully integrated
surface water - groundwater flow models. Environmental Modelling and
Software, 26:886-898.
Partington, D., P. Brunner, C. T. Simmons, A. D. Werner, R. Therrien, G. C.
Dandy, and H. R. Maier. 2012. Evaluation of outputs from automated baseflow
separation methods against simulated baseflow from a physically based, surface
water-groundwater flow model. Journal of Hydrology, 458-459: 28-39.
Partington, D., P. Brunner, S. Frei, C. T. Simmons, R. Therrien, A. D. Werner, H.
R. Maier, G. C. Dandy, J. H. Fleckenstein. Interpreting flow generation
mechanisms from integrated surface water-groundwater flow models of a riparian
wetland and catchment. Submitted to Water Resources Research on 9 November,
2012.
I also give permission for the digital version of my thesis to be made available on
the web, via the University’s digital research repository, the Library catalogue,
and also through web search engines, unless permission has been granted by the
University to restrict access for a period of time.
Signed: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Date: . . . . . . . . .
ix |
ADE | 1. Introduction
Chapter 1
1 Introduction
The understanding of hydrological processes and their translation to the
streamflow hydrograph is critical in the management of floods and water
resources, and the environments they support. A key to this is the understanding
of the interactions of surface and subsurface water systems within catchments
(Winter, 1999; Sophocleous, 2002). The important feedbacks exhibited by these
two systems, necessitate their holistic consideration within catchment hydrology.
Therefore, improving this understanding requires both the identification and
quantification of surface-subsurface water interactions (e.g. losing streams) and
flow generation/depletion processes (e.g. rainfall-runoff and dry-period baseflow).
This means having a clear understanding of the physics of water movement within
catchments. If the physics understanding is clear, and the system is well
characterised, then it follows that the integrated catchment response – in the form
of the streamflow hydrograph – should be able to be readily decomposed into the
constituent flow generation processes, i.e. groundwater discharge and direct
rainfall to the stream, and groundwater discharge and direct rainfall to the
hillslope (see Figure 1.1).
It was highlighted by Hewlett and Troendle (1975) that accurate prediction of the
streamflow hydrograph implies adequate modelling of the sources, flowpaths and
residence time of water. This “adequate” modelling suggests the use of spatially
and temporally distributed hydrological models, of which many have been
developed (see Singh and Woolhiser (2002) for a comprehensive review). It
follows from Hewlett and Troendle’s statement that adequate modelling of the
sources, flowpaths and residence times requires adequate representation of the
physics of water flow, i.e. deterministic-conceptual modelling (see Kampf and
Burges, 2007).
1 |
ADE | 1. Introduction
Figure 1.1: Streamflow generation at the plot scale by both in-stream
(groundwater discharge and rainfall to the stream channel) and overland
(groundwater discharge and rainfall to the hillslope) flow generation
processes.
Freeze and Harlan (1969) provided a blueprint for what is often considered
“adequate” physics-based modelling of water flow within catchments. The
inevitably complex models that arise from this blueprint can aid in building
intuition about the catchment-scale hydrological processes responsible for
streamflow, but subject to the assumptions in the physical equations (e.g. that a
representative elementary volume exists in the subsurface).
In the last decade, the blueprint of Freeze and Harlan (1969) has been realised
with the advent of physics-based fully Integrated Surface-Subsurface
Hydrological Models (ISSHM) (Gaukroger and Werner, 2011). Examples of
ISSHMs include InHM (VanderKwaak and Loague, 2001), MODHMS
(HydroGeoLogic, 2006), HydroGeoSphere (HGS) (Therrien et al., 2009), and
ParFlow (Kollet and Maxwell, 2006). ISSHMs are used within this thesis to
describe models which solve simultaneously the surface and subsurface flow
equations. Within ISSHMs, 2D surface flow is usually represented using an
approximation to the St Venant equations (e.g. diffusion wave), and 3D variably
saturated subsurface flow is usually represented using Richard’s equation.
ISSHMs can be used to analyse and interpret hydrological processes and in
developing conceptual understanding of catchment processes (Ebel and Loague,
2006). A particularly important attribute of these models is that rainfall is
partitioned into infiltration, ponding, and overland flow in a realistic manner
2 |
ADE | 1. Introduction
(Therrien et al., 2009) without any a priori assumption of these processes (Mirus
et al., 2011a). This partitioning is dependent on the rate of rainfall, antecedent
moisture conditions and catchment physical characteristics. However, this means
that the hydrological processes (e.g. groundwater discharge to a stream or
infiltration excess overland flow) need to be identified and interpreted after
simulations.
Studies utilising ISSHMs are becoming increasingly widespread (e.g., Frei et al.,
2010; Maxwell and Kollet, 2008; Park et al., 2011; Brunner et al., 2009). These
examples focused on processes in small-scale synthetic systems, which enabled
insight to be gained into the controls on flow generation (Frei et al., 2010;
Maxwell and Kollet, 2008; Park et al., 2011) and depletion (Brunner et al., 2009).
In larger-scale (e.g. catchment scale) systems it is difficult to resolve how the
hydrological drivers affect the hydrological outputs (e.g. the outlet streamflow
hydrograph). This is because hydrological outputs at a given point in space and
time are only affected by hydrological drivers that occur at the same location and
at the same time (i.e. by ‘active’ processes (Ambroise, 2004)). In larger-scale
systems, hydrological drivers that occur at a particular point in time (active
processes) do not necessarily end up contributing to the hydrological output at that
or a later time. This is because of the influence of travel times, flow impediments
(e.g. riparian wetlands or weirs), and losses (e.g. infiltration or evaporation).
Consequently, where such influences are significant, there is a need to distinguish
between ‘active’ and ‘contributing’ streamflow generation processes (Ambroise,
2004), where contributing processes are those that contribute to flow at a
particular location at a particular time, and potentially include active processes.
These influences will be important in catchments that exhibit significant travel
times for water and/or where flow depletion processes are significant relative to
flow generation processes (e.g. strong losing streams). The differences between
active and contributing flow generation processes are driven by active flow
depletion mechanisms (surface water losses) and the lag-time between active flow
generation mechanisms taking place and the time that the resultant flow reaches
the point of interest (e.g. the point where the streamflow hydrograph is measured).
A key shortcoming of ISSHMs is in linking the distributed hydrologic response to
the point response (e.g. where the streamflow hydrograph is measured), i.e.
3 |
ADE | 1. Introduction
capturing the contributing processes. Active processes are readily obtained from
ISSHMs that output the nodal fluid mass balance components, i.e. surface-
subsurface exchange fluxes, rainfall input, evaporation output, surface inflows and
outflow and changes in storage. However, attaining the contributing processes
requires the ability to use these outputs of active processes to properly analyse and
interpret streamflow generation mechanisms with respect to the streamflow
hydrograph.
The advent of ISSHMs has been critical in being able to improve our conceptual
understanding of hydrological processes, but the benefits of analysing internal
processes and meaningfully separating flow hydrographs are still to be realised.
This is a major shortcoming as it prevents development in building the intuition of
hydrologic response to various hydrological drivers (i.e. rainfall and
evapotranspiration). A clear need has arisen for research into the identification
and quantification of contributing in-stream and overland flow generation
mechanisms at larger (e.g. catchment) scales, particularly given that there are still
difficulties in the ability to conduct or scale up the measurements that are required
in order to gain this understanding at/to the catchment scale (Fleckenstein et al.,
2010).
1.1. Research Objectives
This research aims to improve the understanding of streamflow generation and
surface water-groundwater interaction through quantifying in-stream and overland
flow generation mechanisms within physics-based models of surface-subsurface
flow. In such models, this requires the development of a new method for
interpreting in-stream and overland flow generation mechanisms. This
development will provide a platform for investigation into hydrological systems
that exhibit complex spatiotemporal patterns of in-stream and overland flow
generation mechanisms. To achieve the overall aims of this research, four main
research objectives are developed with three sub-objectives, which are listed
below. The linking of each of these objectives is shown in Figure 1.2.
Objective 1: To develop a method to quantify the contribution of flow generation
mechanisms to streamflow, allowing separation of the streamflow hydrograph into
its constituent flow generation components (i.e. groundwater discharge and direct
4 |
ADE | 1. Introduction
Quantifying flow generation
mechanisms1
Development of method to quantify Implementation of
streamflow generation mechanisms1 method within an
ISSHM5
In-stream flow generation Overland flow generation
mechanisms1.1, 1.2 mechanisms1.2
Groundwater Direct Overland Streamflow Direct Groundwater
discharge to rainfall on flow to to rainfall on discharge to
stream1.1 ,1.2 stream1.1, 1.2 stream1.2, 2 overland1.2 overland1.2 overland1.2
Baseflow
benchmark2
Separated
Hydrographs1
Spatiotemporal Active vs.
Analysis3 Contributing4
Testing automated
baseflow separation
methods2.2
Figure 1.2: Research objectives and their hierarchy. Objectives are denoted
by the superscript numbers in each of the flowchart boxes.
1.2. Thesis Overview
This thesis is organised into five chapters. The main body of this thesis consists of
Chapters 2 to 4, which correspond to three journal papers (Partington et al.,
2011; Partington et al., 2012a; Partington et al., 2012b). In Chapter 2
(Partington et al., 2011) a new method is developed for accurately quantifying
groundwater contributions to streamflow (Objective 1) with respect to the
streamflow hydrograph, and this method is used to investigate the complexity of
groundwater contributions to streamflow (Objective 1.1). In Chapter 3
(Partington et al., 2012a) the work in Chapter 2 is extended, and a baseflow
model benchmark is developed (Objective 2) against which potential error is
investigated in commonly used automated methods for the separation of baseflow
from streamflow hydrographs (Objective 2.1). In Chapter 4 (Partington et al.,
2012b) the work of Chapter 2 is extended to investigate overland flow generation
mechanisms (Objective 1.2), and the new method is applied to a case study of a
real catchment. Within the case study, all surface flow generation mechanisms are
quantified (Objective 1.2), spatiotemporal variability in flow generation
mechanisms is analysed (Objective 3), and the difference between active and
6 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
Abstract
The complexity of available hydrological models continues to increase, with fully
integrated surface water-groundwater flow and transport models now available.
Nevertheless, an accurate quantification of streamflow generation mechanisms
within these models is not yet possible. For example, such models do not report
the groundwater component of streamflow at a particular point along the stream.
Instead, the groundwater component of streamflow is approximated either from
tracer transport simulations or by the sum of exchange fluxes between the surface
and the subsurface along the river. In this study, a hydraulic mixing-cell (HMC)
method is developed and tested that allows to accurately determine the
groundwater component of streamflow by using only the flow solution from fully
integrated surface water - groundwater flow models. By using the HMC method,
the groundwater component of streamflow can be extracted accurately at any
point along a stream provided the subsurface/surface exchanges along the stream
are calculated by the model. A key advantage of the HMC method is that only
hydraulic information is used, thus the simulation of tracer transport is not
required. Two numerical experiments are presented, the first to test the HMC
method and the second to demonstrate that it quantifies the groundwater
component of streamflow accurately.
2.1. Introduction
A quantitative understanding of stream flow hydrographs is an important
precondition to the understanding and effective management of any catchment
(VanderKwaak and Loague, 2001; Jones et al., 2006; Mirus et al., 2009). The
streamflow hydrograph is generated by different mechanisms such as groundwater
discharge to the stream, discharge from the unsaturated zone, overland flow,
preferential flow through macropores and/or fractures, and direct precipitation to
the stream. These streamflow generation components can exhibit complex spatial
and temporal behaviour. This complexity makes it difficult to easily decompose
stream flow hydrographs in terms of stream flow generation mechanisms if one or
several components of the hydrograph are unknown. Groundwater discharge is a
critical streamflow generation component that is difficult to quantify. The
quantitative assessment of the groundwater component of streamflow (which
13 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
represents the quantity of streamflow at a given point in space and time consisting
of groundwater discharging directly to the stream) is of great importance in
understanding catchment hydrology and informing water resources management,
as highlighted by Sophocleous (2002) and Winter (1999). Accurate simulation of
the groundwater component of streamflow is therefore important in hydrological
modelling exercises (e.g. Gilfedder et al., 2009; Croton and Barry, 2001; Facchi
et al., 2004) in order to inform water resources management.
The groundwater component of stream flow cannot be measured easily in the field
(Hatterman et al., 2004) and therefore is usually quantified using indirect
methods. Indirect methods can involve the use of environmental and conservative
tracers for separation of the hydrograph (McGlynn and McDonnell, 2003;
McGuire and McDonnell, 2006), and recession analysis based on conceptual
storage-discharge relationships for the catchment (Chapman, 2003; Eckhardt,
2008). However, as pointed out by Hewlett and Troendle (1975), ‘the accurate
prediction of the hydrograph implies adequate modelling of the sources, flowpaths
and residence time of water’. In particular, capturing the flowpaths requires a
spatially distributed model. Unless the assumptions of the indirect methods can be
resolved or justified, the adequate modelling of sources and flowpaths of water
would be insufficient. If the modelling is insufficient, then it follows that the
separation of the hydrograph may be meaningless. Given the difficulty faced in
accurately measuring sources and flowpaths within hillslopes, let alone entire
catchments, some benefit can be found in examining hypotheses which can be
adequately ‘measured’ in the ‘virtual laboratory’ (Weiler and McDonnell, 2006).
One could expect that the tools for quantifying the groundwater component of
streamflow are now readily available in the latest generation of fully integrated
spatially distributed models such as InHM (VanderKwaak and Loague, 2001),
MODHMS (HydroGeoLogic, 2006), HydroGeoSphere (HGS) (Therrien et al.,
2009), Wash123D (Cheng et al., 2005) and ParFlow (Kollet and Maxwell, 2006).
However, this is not the case. Even within spatially distributed numerical models
quantifying source components remains a challenge (Sayama and McDonnell,
2009). The same applies to the ultimate delivery mechanisms as defined in Sklash
and Farvolden (1979). Because the currently available numerical models do not
14 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
report the groundwater component of streamflow at a given location, it is often
approximated by introducing tracers or by setting it equal to the summed
exfiltration along a section or entire length of the stream. The summed exfiltration
is defined in this paper as the sum of all fluxes from the subsurface to the stream
at a specific point in time upstream of the point at which the hydrograph is
measured.
However, these approaches are problematic. For example, the summed exfiltration
during a simulation is not equal to the groundwater component of streamflow at
the same simulation time. This can be attributed to the fact that portions of the
summed exfiltration exhibit a time lag from the point of entering the stream to the
point of streamflow measurement, as a result of potentially significant transit
times within stream networks (McGuire and McDonnell, 2006). This time lag
cannot be captured if the groundwater component of streamflow is approximated
by the summed exfiltration. Furthermore, if the stream loses water to the
subsurface between a point of groundwater discharging into the stream and the
point where the hydrograph is measured, only a portion of the groundwater
entering the stream will contribute to the groundwater component of streamflow
at the point of hydrograph measurement. In that case, the summed exfiltration will
overestimate the groundwater component of streamflow at the point of
hydrograph measurement.
In this study, a mixing-cell method for quantifying the groundwater component of
streamflow in fully integrated spatially distributed models is described. Mixing-
cell models have often been used in hydrogeology to model solute transport (Adar
et al., 1988; Campana and Simpson, 1984). Mixing-cell models rely only on
conservation of mass. The hydraulic mixing-cell (HMC) method described in this
study relies on hydraulic information only (i.e. fluxes). Moreover, the method
allows tracking streamflow generation mechanisms at every cell or element within
the stream of the model domain. Therefore, complex spatial and temporal effects
are captured and can be accounted for. The method is developed and tested using
a particular numerical model (HydroGeoShpere, Therrien et al., 2009), but it can
be implemented to any code that reports the exchange between the subsurface and
surface in a spatially distributed manner. The paper also aims to explore the
15 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
suitability of traditional methods (e.g. equilibrating the groundwater component of
streamflow to the summed exfiltration) for quantifying the groundwater
component of streamflow within numerical models.
2.2. Existing methods for extracting streamflow generation
components
The hypothetical catchment shown in Figure 2.1 is used to illustrate the
challenges of extracting the groundwater component of streamflow from
numerical models using existing methods. In the catchment shown, the stream,
which is flowing from A to B to C, is gaining in sections A and C, but losing in
section B.
Figure 2.1: Conceptual diagram of a surface water-groundwater catchment
(left hand side) featuring different flow regimes (as illustrated in the right
part of the figure). The white sections of the catchment adjacent to the
stream represent the groundwater discharge upslope of the stream (return
flow). The dashed lines on the right part of the figure represent the water
table. The flow direction is towards the reader.
2.2.1. Summed exfiltration along the length of the stream
For each of cross sections A, B and C of the hypothetical catchment shown in
Figure 2.1, the expected streamflow hydrograph is shown in Figure 2.2, along
with the groundwater component of streamflow and the summed exfiltration. The
streamflow in Figure 2.2 A, B and C refers to the point measurement at each of
cross sections A, B and C. Although the results shown in Figure 2.2 are
hypothetical, they illustrate the following two problems that arise by
16 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
approximating the groundwater component of the streamflow using the summed
exfiltration:
1) the summed exfiltration does not account for the time lag between the upstream
points of groundwater discharging from the aquifer to the stream and the point
where the hydrograph is measured, as illustrated by the time lag between the
summed exfiltration and the groundwater component of streamflow curves. The
streamflow travel times for the summed exfiltration upstream of cross sections A,
B and C actually correspond to the time lag between the peaks of the summed
exfiltration and the streamflow hydrograph in Figure 2.2.
2) changing flow regimes cannot be considered correctly. When a part of the
stream is losing and other parts are gaining, the summed exfiltration is not equal
to the groundwater component of streamflow at a particular location, even if the
aforementioned time lag is negligible.
The effect of ignoring time lags and discounting losses along the stream becomes
clear when moving downstream from cross sections A to B to C. For example, the
course of the groundwater component of streamflow at cross section A features a
flatter and broader distribution through time compared to the summed exfiltration
upstream of A. When considering the streamflow hydrograph at cross section C in
Figure 2.2, the significance of time lags, particularly from the most upstream sub-
catchments, becomes apparent.
Figure 2.2: Hydrograph at cross sections A, B and C of the catchment shown
in Figure 2.1. The streamflow and corresponding component of groundwater
flowing through cross sections A, B and C are shown. Also, the summed
exfiltration upstream of cross sections A, B and C, respectively, are shown.
2.2.2. Tracer based hydrograph separation
The use of conservative tracers within models provides temporal information on
the original source of water (i.e. groundwater, soil water, rainfall). Whilst the
17 |
ADE | 2. A hydraulic mixing-cell method to quantify the groundwater component of
streamflow within spatially distributed fully integrated surface water -
groundwater flow models. (Paper 1)
application of solutes is extremely useful in identifying the source of streamflow,
it gives no real indication of the mechanism of streamflow generation (McGuire
and McDonnell, 2006). Even with temporal information on the source of water,
the parameters associated with tracer transport (i.e. diffusion, tortuosity and
dispersivity) often affect the interpretation of the source as demonstrated in Jones
et al. (2006). Jones et al. (2006) found that the value of dispersivity used in
simulating the transport of tracers could lead to large overestimation of the pre-
event water’s contribution to streamflow. In their model using InHM of the
Borden rainfall-runoff experiment, the pre-event contributions to streamflow
using longitudinal dispersion (cid:2) = 0.5 m and 0.005 m were found to be 41.6% and
L
33.9%, respectively, with the hydraulically based subsurface contribution close to
0%. These results would suggest that in the streamflow hydrograph in Figure 2.1
at cross section C of the catchment, the groundwater component of streamflow
could be easily overestimated using tracers as illustrated in Figure 2.3.
Figure 2.3: The theoretical hydrograph at cross section C of the catchment
shown in Figure 2.1. The streamflow, groundwater discharge component and
tracer based separation (for dispersivity values of (cid:2) and (cid:2) ) are shown.
L1 L2
Given such large variation in the tracer based interpretations of groundwater
contributions to streamflow, it seems quite clear that inherent accuracy relies on
reliability and certainty of the transport parameters. Any uncertainty in the
dispersivity directly relates to uncertainty in quantifying the groundwater
component of streamflow. Therefore quantifying the groundwater component of
the streamflow hydrograph within models using tracers may be undermined by
large uncertainty.
18 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.