I am grateful to the Board of the University of Wageningen for giving me the permission to pursue my study ... international students, Ms. Jennine Herman and the coordinator of the ...... ensicM-mmy.fr/COSTWWTP/Benchmark/Benchnuirkl}itm).
DevelopmentofaBenchmarkingMethodologyfor EvaluatingOxidationDitchControlStrategies
AbdallaA.A.Abusam
Promotor:
prof.dr.ir.G.vanStraten HoogleraarindeMeet-,Regel-enSysteemtechniek
Co-promotoren:
dr.ir.KJ. Keesman Universitairhoofddocent, leerstoelgroepMeet-,Regel-enSysteemtechniek dr.ir. H. Spanjers Universitairdocent,sectieMilieutechnologie
Samenstellingpromotiecommissie: prof.Dr.ir.P.A.O.Vanrolleghem(Universiteit Gent) prof.Dr.ir.W.H.Rulkens(Wageningen Universiteit) prof.Dr.ir.J.Grasman(Wageningen Universiteit) dr.ir.S.R.Weijers (TechnisheUniversiteitEindhoven)
^AJono! t^O^i
AbdallaAbdelgadirAhmedAbusam
DevelopmentofaBenchmarkingMethodologyfor EvaluatingOxidationDitchControlStrategies
PROEFSCHRIFT terverkrijgingvandegraadvandoctor opgezagvanderector magnificus vanWageningenUniversiteit, prof.dr.ir.L.Speelman, inhetopenbaarteverdedigen opmaandag 17September2001 desnamiddagstehalftweeindeAula
U
>\ \ 6? r-og)^
CIP-DataKoninlclijkeBiblioteek.DENHAAG Abusam.A.A.A. Development of a Benchmarking Methodology for Evaluating Oxidation Ditch Control Strategies/ A.A.A.Abusam [S.I.:s.n.] ThesisWageningenUniversity.- Withref. - WithsummaryinDutch ISBN90-5808-422-1
MKioSZO \
%3oZ&.
Propositionsattachedtothethesis: "Developmentofabenchmarkingmethodologyforevaluatingoxidation ditchcontrolstrategies",byA.A.A.Abusam 1.Oxidationditcheshavemanyadvantagesoverotheractivated sludgesystems. 2."Inmodernwatermanagementmodelsareindispensable." R.H.vanWaveren(1999),Wat. Sci.Tech.39(4):13-20. 3.10to ISCSTR's areadequate for modelingtheeffluent qualityofanoxidation ditch. 4. The response surface method (RSM)leadsin a systematicway to good initial guesses ofcomplexsystemparameters. 5. Any horizontal (recirculation) velocity in the range of 0.25 to 0.60 m/s is often recommended to prevent settling of organic matters to the channel bottom, however, variations of the horizontal velocity within this range significantly affect the nitrogen removalprocess inoxidation ditches. 6. Influent step feeding will not significantly influence the internal distribution of the sludgeinoxidationditches. 7. Although it is rarely carried out,backward uncertainty analysis results in very useful information. 8."Employingavacuumflushintoiletsandseparatingthesetoiletflushesatsourcefrom the remaining grey sewage and urban surface runoff is sufficient to open a myriad combinatorialpossibilitiesfor thecomposition ofawastewater infrastructure'' M.B.Beck(1999),Wat.Sci.Tech.39(4):1-11. 9.Futureisfor decentralized andindividualhouseholdwastewatertreatmentsystems. 10.In the near future, the ethical ramification of research onhuman cloning will be one ofthemostcontroversialissues. 11. For somethis isthe 3 rdmillennium, for the Egyptians it isthe 7thmillennium,for the Mayansitisthe9*millennium,forsomeothersitismorethan that
Abstract Abusam,A.A.A. (2001),Developmentofabenchmarkingmethodology forevaluatingoxidationditchcontrol strategies,Ph.D.thesis,WageningenUniversity,Wageningen,TheNetherlands. Thepurpose of thisthesis was todevelop abenchmarking methodology forevaluating control strategies for oxidationditchwastewatertreatmentplants.Abenchmarkconsistsofadescriptionoftheplantlayout,abasic simulation model (reactor, settler, sensors andactuators models) anddefinitions of (controller) performance criteria. The goal was achieved by outlining theprocedure for developing such a benchmark for a specific full-scale WWTP,usingavailableprocessdata.ForotherWWTP's,thesameprocedurecanbefollowed. Indeveloping thebasic simulationmodel, firstaloop-of-GS7R's model,withoutback-flows, waschosenfor modelingoxidationditchesbecause itissimpleandcanbeusedforcontrolpurposes.Basedonthismodel,a new method for estimating the standard oxygen transfer rate (SOTR),under clean water conditions, was developedandtestedsuccessfully. ThenewmethodestimatestheSOTR onthebasesoftheaerationconstant, which is the product of KLa and volume of aerated compartment, because neither KLa nor the volume of aeratedcompartment can be individually identified. Underprocess conditions, C-oxidationand nitrification processeswereassumedtotakeplaceintheaeratedzones,whereasthedenitrificationprocesswasassumedto occur in the anoxic zones. For modeling the biochemical processes, ASM No. 1 was used, whereas for modeling the secondary settlerthenon-reactive double-exponential settlingvelocity model was used. Based oninfluent-effluent concentrations, itwasfoundthathydraulics ofoxidation ditchescanbeapproximatedby 10 to IS CSTR's.The oxidation ditch model was then calibrated successfully using a novel calibration strategy, which is based on response surface analysis. Prior to a formal parameter estimation step, the responsesurfaceanalysisprovidesinsightintheparametersensitivity andinitialestimates.Becausethestudy was limitedtoCandNremovalprocesses,only modelsofDOandNsensorsweredeveloped. Theactuators, pumpsandvalves,wereassumedtoworkperfectly, thatis:dynamicsandtimedelaysoftheseactuatorswere neglected. Evaluation criteria were then developed by modifying the criteria proposed by both COST 624 Working GroupandIWATaskGrouponRespirometry.Modifications weremainlymadeintheaerationandpumping energy equations, because oxidation ditches use mechanical aerators that are different from air diffusers adopted by COST 624 and IWA Working Groups. In addition, long-term evaluation criteria were also developed. Further, sensitivity analysis was carried out to determine parameters of ASM No. 1 that require special attentionfromthebenchmarkuser. Sensitivity analysiswascarriedoutusingthefactorial sensitivity analysis methodology. Themainadvantage ofthismethodology is thatmoreinformation abouttheinteractions(nonlinearities) canbeobtained. Also,theeffect ofthevarioussources ofuncertainty ontheperformance indices wasinvestigated. Estimationoftheuncertainty contributionof thevarioussources is very importantbecause itenablesthebenchmarkusertomakeanappropriateselectionamongdifferent controlstrategies.Itisequally importantfordesigningexperimentalormonitoringprogramswiththeaimofreducingtheuncertainty. Finally,thebenchmarking procedurewasdescribed anddemonstratedbyusing itto evaluatesomebasicand advanced control strategies. Basic control strategies studied were (i) splitting the influent flow between the aeratedcompartments,(ii)rateofactivatedsludgerecirculatedand(iii)aerationpatterns.Thebenchmarkwas also used in studying the effect of the horizontal (recirculation) velocity onnitrogenremoval process.Here, the horizontal velocity was considered as a manipulated control variable, to obtain maximum 77Vremoval efficiency. Keywords; wastewater, oxidation ditch, carrousel, modeling, activated sludge,ASMNo. 1,oxygen transfer rate,aeration,parameter estimation, calibration, sensitivity analysis,uncertainty analysis, sensors,horizontal velocity,benchmark,benchmarking,controlstrategies,simulation.
Tothememoryofmyparents
Acknowledgements Itisagreatpleasure that nowIhavethechancetothank thosewhohave contributed tothisthesis. FirstIwouldliketoexpressmysinceregratitudetomypromoter,Prof.Dr.jr.GerritvanStraten,for givingmetheopportunity topursue mydoctoral studiesattheSystemsandControl Group andfor the wonderful guidance, constructive discussions, valuable suggestions andcritical readings ofmy work. My deep appreciations are also due to Dr. ir. Karel J. Keesman, my co-promotor and daily supervisor, for choosing the benchmarking problem for me. I also thank him for the skillful guidanceandsuggestions,forthepromptadvicewheneverIcontacted him,andforthe constructive discussionsofmywork-plans.Asamatteroffact,Iamalsograteful toDr.ir.TonvanBoxtel,who introducedmetoDr.ir.KarelKeesman. My grateful thanks are also due to Dr. ir. Henry Spanjers, my co-promotor, for all the joyful discussionswehad, forhisthorough commentsovermywritings andhissharpcriticismswith"the rightwordattherightplace",forthevaluablehelpinreachingtherelevant literature,especiallythe Dutch literature, andforthecontactshemade formewith anumber oftreatment plants andwater authorities. I amdeeply indebted to ir.Kees Meinema,fromDHVWater, forthediscussions wehad,forhis contributiontothewritingofmostofthepapers,forcheckingthebenchmarking procedure,andfor thepracticalexperienceshehasputintomywork. I gratefully acknowledge thefruitful discussions byDrJohn B.Copp,thecoordinator oftheIWA Task Group onRespirometery, whoused toattend ourmeetings during hisstay in Wageningen. I also thank him for his willingness to answer my questions even after he has returned back to Canada,throughthee-mail. I amgrateful totheBoard oftheUniversity ofWageningen forgivingmethepermissiontopursue my study at Wageningen University. In particular, I would like to thanks the former Dean for international students, Ms.Jennine Herman andthe coordinator of the environmental studies, ir. DickLegger. IalsogreatlyappreciatethehelpgivenbyDHVWaterintheNetherlands,intermsofrealdataused inthisstudyandfinancial support forpresentingapaperattheWatermatex2000 conference heldin Gent, Belgium. In particular, I would like to thank B.P.A. Hoitink and E.F.J. van Heijden who helpedmetocollectthedata. I would like also to thank ir. F.T. van Breukelen of Hoogheemraadschap van Schieland, for providingmewithextrarealdataofthefull-scale wastewatertreatmentplantthatIhavestudied. I alsothankAddeManofZuiveringschap Limburgforgivingmethechancetopresentanddiscuss thethesisfindingswiththeDutchandFlemishexperts. I also would like to thank Prof. Dr.ir.P.A.Vanrollegem from theDepartment ofBiomath., Gent University,Belgium,forthediscussions,encouragementsandforsendingmerealdatathatIusedto testthemodelIproposedfortheseasonaleffects oftemperature.
I thank my Mexican office-mate, Irineo Lopez, for the nice time we have spent either working, discussing orjust talking.I alsothankhimfor helpingmeto solve someofthecomputerproblems. In the name of my family, I also would like to thank his family for the friendly relationship they havewithmy family. I alsowould like tothank all the staff of Systems and Control Group for thefriendlytreatment. In particular,Iwouldliketothankthecomputersystemmanager,HenkVeen,for solvingmycomputer troubles.Ialso liketothank the "De feestcommissie": Gerard vanwillenburg,Usequirijns,Camile Hoi (recently moved back to Delft Technical University) and Stephan de Graaf who used to encourageus (meand Irineo)to participate inthe social activitiestaking place inthedepartment. I also thank Camile Hoi for making a really nice Dutch Summary (Samenvatting) out of my translationtrials. Lastlybutnot least,Iwould liketothank mywifeNagat and mydaughter Razan for the continued supportandencouragements.Ialsoliketothankmybrothers,sisters,relativesandfriends inSudan, USAandtheNetherlandsforthecontinuedencouragements.
Tableofcontents IV
Abstract Dedication Acknowledgements
V
1.Introduction 1.1 Background 1.1.1 Biologicalwastewatertreatmentplants 1.1.2 Oxidationditches 1.1.3NeedforadvancedcontrolsystemsinbiologicalWWTP's 1.2 Defining thebenchmarkingproblem 1.3Research objectives 1.3.1 General objective 1.3.2 Specific objectives 1.4Contributionsofthethesis 1.5Research methodology 1.6Researchfocusandlimitations 1.7Outlineofthethesis 1.8 References
1 1 1 2 5 6 6 6 7 7 7 8 8 9
vi
PART1,MODELING 2.Oxygentransferrateestimationinoxidationditches fromcleanwatermeasurements 2.1Abstract 2.2Introduction 2.3Theproposedmethod 2.4Applicationtorealdata 2.4.1Briefdescription oftheplant 2.4.2BriefdescriptionofSOTRmeasurements 2.5Resultsanddiscussionofthenewmethod 2.5.1Modellingtheoxidationditch 2.5.2EstimationofKLAandVA 2.5.3EstimationofkandVA 2.5.4Estimationofkonly 2.5.5EstimationofSOTR 2.6Conclusions 2.7References 2.8Appendix,MainequationsSTORAmethod 3.Effect ofnumberofCSTR'sonthemodellingofoxidationditches:steady state anddynamic analysis 3.1Abstract 3.2Introduction 3.3Materialsandmethods 3.3.1Plantlayout 3.3.2 Simulationmodel
14 15 16 18 18 19 19 19 20 21 21 22 23 24 24
25 25 26 26 26
3.3.3Methods 3.4Resultsanddiscussion 3.4.1Steadystatesimulations 3.4.2Dynamicsimulations 3.5Conclusions 3.6References 4.Parameterestimationprocedureforcomplexnon-linearsystems: calibrationofASMNo.l forN-removalinafull-scaleoxidation ditch 4.1Abstract 4.2Introduction 4.3Proposedprocedure 4.3.1 Step 1,Specify plantlayoutandmodel 4.3.2Step2,Collectin/outputandoperationaldata 4.3.3Step3,Estimateinitialstateconditionsfrompastdata 4.3.4Step4,UseRSMtoselectdominantparameters 4.3.5 Step5,Applyformalparameterestimationmethod 4.3.6step6,Evaluateestimationresults 4.4Conclusions 4.5References 4.6AppendixA:Operationalpatternoftheaerators 4.7AppendixB:Directionsofdominantparameters 4.8AppendixC:Evaluationofparameteruncertainty
26 27 27 28 29 29
31 31 32 33 33 34 34 37 38 39 39 40 41 41
PART2,MODELANALYSIS 5.Sensitivityanalysisonoxidationditches:theeffect ofvariationsinstoichiometric, kineticandoperatingparametersontheperformance indices 5.1Abstract 5.2Introduction 5.3Performance indices 5.4Method 5.5Resultsanddiscussion 5.5.1Firststage:Parametersmaineffects 5.5.2Secondstage:Factorialsensitivityanalysis 5.6Conclusions 5.7References
43 45 46 47 49 49 50 54 55
6.Uncertainty analysis 6.1Estimationofuncertaintiesintheperformanceindicesofanoxidationditch benchmark 6.1.1Abstract 6.1.2Introduction 6.1.3Theory 6.1.4Performance indices 6.1.5Estimationofperformance indicesuncertainties 6.1.5.1 Parametervaluesandinputloadsuncertainties 6.1.5.1.1 Method 6.1.5.1.2Resultsanddiscussion
57 58 59 60 62 62 62 63
6.1.5.2Initialstateuncertainties 6.1.5.2.1Methods 6.1.S.2.2Resultsanddiscussion 6.1.5.3Modelstructureuncertainties 6.1.5.3.1Methods 6.1.5.3.2Resultsanddiscussion 6.1.5.4Uncertaintiesduetoseasonalchangesinwatertemperature 6.1.5.4.1Methods 6.1.5.4.2 Resultsanddiscussion 6.1.6General discussion 6.1.7 Conclusions 6.1.8 References
66 66 67 67 67 68 69 69 71 72 72 73
6.2Forwardandbackwarduncertaintypropagationinmathematical models 6.2.1Abstract 6.2.2 Introduction 6.2.3Theory 6.2.4Workingexample 6.2.5 Conclusions 6.2.6 References
75 75 76 77 83 83
PART3,BENCHMARKING 7.Implementationofthebenchmark 7.1Introduction 7.2Componentsofthebenchmark 7.3Step-by-stepbenchmarkingprocedure 7.4References 8.Illustration oftheuseofthebenchmark 8.1Introduction 8.2Effect ofthehorizontalvelocityonperformance ofoxidationditches 8.2.1 Abstract 8.2.2 Introduction 8.2.3 Simulationexample 8.2.4Practicalassessment 8.2.5 Conclusions 8.2.6 References 8 3 Evaluationofcontrolstrategiesusedinoxidationditchplants 8.3.1Abstract 8.3.2 Introduction 8.3.3 Benchmarkingaspecific WWTP 8.3.3.1Plantlayout 8.3.3.2Modeldevelopmentandvalidation 8.3.3.3Performance criteria 8.3.3.4Testingofthebenchmark 8.3.4 Implementationofthebenchmark 8.3.4.1Evaluationofshort-termcontrolstrategies
86 86 86 91 93 94 94 94 95 98 100 100 102 102 103 103 103 105 105 105 106
8.3.4.1.1ControlstrategyNo. 1,splittingoftheinfluent 8.3.4.1.2ControlstrategyNo.2,RAS 8.3.4.1.3ControlstrategyNo.3,aerationpatterns 8.3.4.2Evaluationoflong-termcontrolstrategies 8.3.5 Conclusions 8.3.6 References 9.Generaldiscussionandconclusions 9.1Discussion 9.2Generalconclusions AppendixI,ASMNo.1 AppendixII,Performance indices AppendixIII,Modellingofsensorsandactuators AppendixIV,Descriptionofthebenchmarkedfull-scale WWTP AppendixV,Matlab/Simulinkmodelofthefull-scaleWWTP
flow
106 106 107 108 111 111 113 115 118 119 127 132 134
Summary Samenvatting
143 146
CurriculumVitae
149
1.Introduction 1.1 Background 1.1.1Biologicalwastewatertreatmentplants Collection andtreatmentofwastewater disposals isgenerallyassociated withthegrowth ofcities,whichresultedfromtheindustrial revolution (Fairand Geyer, 1958).Inthe19th century, after the Great Plague in London, use of conventional biological wastewater treatment plants (WWTP's) has started. At that time, the objectives of wastewater treatment were mainly concentrated on basic public health issues such as prevention of epidemics,protection ofsources ofpotablewater,and prevention ofnuisance conditions likeproduction of nasty odours and presence ofvectors (Lester, 1996).However, inthe pastcentury,theobjectives havesomewhatchanged(Metcalf&Eddy, 1991).Fromabout 1900 to 1970, the objectives of wastewater treatment were: (i) removal of suspended solidsand floatable matters,(ii)treatment ofbiodegradableorganics and (iii) elimination ofpathogenic organisms.Fromabout 1970to 1980,theobjectives weremainlybased on aesthetic and environmental concerns. From 1980 up to now, the main concerns are removalofnutrients,likenitrogenandphosphorus,whichmaycauseeutrophication. Conventionalwastewatertreatmentisacombination ofphysicalandbiologicalprocesses. It consists of the following four steps: (i) preliminary treatment: screening and grit removal (ii) primary treatment (this step is often not practised in The Netherlands): removal of 30-50% of the suspended solids (TSS) in a primary settling tank, (iii) secondary treatment: biological treatment, which is usually a trickling filter or an activated sludge reactor, (iv) advanced treatment: in some conventional WWTP's chlorinedisinfection isapplied priortodischargetoareceivingwater.Althoughtheyare very efficient in removing suspended solids and organic matter (more than 85percent), conventional WWTP's are usually very poor in removing nitrogen, phosphorus, heavy metals, nonbiodegradable organics, bacteria and viruses (Qasim, 1999). Regarding nutrients removals in conventional WWTP's,totalnitrogen (IN) removal isabout25-55 percent,whereastotalphosphorus(TP) removal isabout10-30percent(OrhonandArtan, 1994). Inthe lastfew decades,knowledgeandpublicawareness aboutwaterpollution problems have significantly increased. Due to that, environmental problems, such as: eutrophication, depletion of oxygen and toxicity to fish, have been associated with discharges of WWTP's into the receiving waters. It has been found that elimination of only the organic matter in discharges of WWTP's will not prevent the eutrophication problem,asnitrogenandphosphorus stillcansupportbiomassgrowth(Orhon and Artan, 1994). Hence advanced treatment processes like biological nitrogen removal and biological/chemicalphosphorusremovalhavebeenintroduced inWWTP's. Biological nitrogen removal is a two-step process: nitrification followed by denitrification. Inthenitrification processammoniaandorganicnitrogenareconvertedto nitrate,whereas inthe denitrification processnitrateisconvertedtonitrogen gas.Table1
presentsthetypical reactions oforganicmatteroxidation andnitrogenremoval processes thattakeplaceinbiologicalWWTP's.
Table 1,TypicalreactionsinbiologicalWWTP's (RittmanandLangeland,1985). Aerobic,heterotrophicoxidationoforganicmatter C5H9ON+J02+0.4H+ ->2C02 +0.6C5HTO2N+0ANHZ +l.8H20 Aerobic,autotrophicoxidationofammonium Mf4++1.802+0.2CO2 ->0.96M>3+0.04C5H1O2N+\.64H+ Anoxic,heterotrophicdenitrification oforganicmatter C5H,ON+336NO; +3.92H+ -+l.6$N2 +036C5H7O2N+3.2CO2+3.92H2O+0.64NHt Note, C5H9ON representsorganicmatter. CsH-p2N representsbacteria. In comparisontophysical orchemical removalprocesses,biological removalofnitrogen has the following mainadvantages: (i)moderate costs, (ii) high removal efficiency, and (iii)highprocessstabilityandreliability (Metcalf&Eddy, 1991).Forachieving efficient biological nitrogen removal, many reactor configurations have been proposed. These configurations can broadly be classified as single-sludge systems and multi-sludge systems, hi a single-sludge system, simultaneous carbon oxidation, nitrification and denitrification processes are accomplished with the same sludge, whereas in a multisludge system, the nitrification and denitrification processes occur in separate reactors with different biomass populations. However, both single-and multi-sludge systemscan further besubdivided intopre-denitrification systems,wherenoexternalcarbonsourceis used, and post-denitrification systems, where external carbon source is added to the system. Oxidation ditches and sequential batch reactors (SBR's) are good examples of single-sludgesystems.
1.1.2 Oxidationditches Oxidation ditches are variants of the activated sludge system. They are single-sludge wastewater treatment systems. This means that they are capable of achieving carbon oxidation, nitrification and denitrification in single biomass slurry. Due to presence of aerobic, anoxic and anaerobic zones, in fact, oxidation ditches are capable of achieving not only C and N removals, but also P removal (WEF, 1998). However, this thesis focusesonCandNremovals,only. Since its original application in The Netherlands, the oxidation ditch has become a significant wastewater treatment technique all over the world (Huang and Drew, 198S). hi the Netherlands, me oxidation ditch is the most widely used wastewater treatment system (CBS, 2000). An oxidation ditch continuously recirculates the mixed liquor
through a closed-loop, oval-shaped channel equipped with mechanical aerators (rotors) that are usually placed in series along the channel. Mechanical aerators are used to introduce oxygen into the system, and to provide sufficient horizontal velocity that prevents the organic solids from settling in the channel bottom surface. Typically, the horizontalvelocityisbetween0.25and0.35m/s(Metcalf&Eddy,1991). The original version of oxidation ditches, as developed by Pasveer in the 1950s in The Netherlands, has a channel depth of about 1.5 m, equipped with brush aerators, and internal secondary settler (Pasveer, 1971). This first version of oxidation ditches is basically developed for use ina small community. Foruse ina larger community (e.g.> 100000p.e.),thistypeofoxidationditchisfoundtobeeconomicallynotfeasible.Dueto the shallow depth (1.5 m), a large surface area and a high number of aerators that are required. In 1968,DHVWaterBVhasdeveloped anewversion ofoxidation ditches,theso-called carrousel oxidation ditch, which is economically a feasible system for use in large communities ofupto 500000p.e. (KootandZepers, 1972).Thecarrouselchannel hasa depthof4to5m.Thusitoccupieslesssurfaceareathantheoriginalversionofoxidation ditches developed by Pasveer in the 1950's. Further, the deep channel of the carrousel allows installation of vertically mounted aerators that are more efficient than the horizontally mounted brush rotors. In addition, vertically mounted mechanical aerators add more control flexibility to the system, as various operating settings can easily be achieved by changing the number of aerators in use, the rotational speed, and/or the immersiondepth. At almost the same time as the carrousel, the orbal system (a multi-channel oxidation ditch system) wasalsodeveloped in SouthAfrica (Drews era/., 1972).Theorbal system consists of a number of concentric oval aeration channels connected in series, followed by a secondary settler. The settler is actually surrounded by the aeration channels. The aeration mechanism ofthe orbal system consists ofa number ofperforated disks,which are partly immersed and rotate around horizontal axes. Increasing or decreasing the numberofthediscscontrolsoxygeninputintotheorbalsystem. Recently, DHV Water BV has proposed a new version of the carrousel, the so-called carrousel-2000.TNremovalefficiency ofthissystemisexpectedtobehigherthanthatof the other oxidation ditch systems, as a separated denitrification compartment will be allocated within the ditch (DHV Water, 1993). A similar modification has also been suggested by Sen et al, (1992), who proposed aerating only the first half of the ditch while leaving the second half anoxic, in order to achieve high TN and TP removal efficiencies. In small oxidation ditches (e.g. under 4000 p.e.), intermittent aeration has alsobeensuccessfully appliedforachievinghigh TNremoval(Inomaeetal, 1987;Araki etal, 1990). In comparison to other activated sludge systems, oxidation ditches have many advantages. First, alternating aerobic and anoxic zones exist along the ditch, due to the location of the aerators in series along the ditch channel. Consequently, simultaneous
removal of organicandnitrogenous matteroccurs repeatedly in oxidation ditches.Given thatonly 10to30minutesareusuallyneededfor recirculatingthewastewateraroundthe ditch,biomassundergoesarapid changeofaerobicandanoxicconditions,which,inturn, stimulates growth of various types of microorganism in the oxidation ditches. Heterotrophic and autotrophic bateria grow in the aerated zones, whereas denitrifying bacteria grow inthe anoxiczones. Therefore, C-oxidation and nitrification takeplace in the aerated zones, whereas the denitrification process occurs in the anoxic zone. However, this is not the only explanation for the efficient denitrification process occurring in oxidation ditches.Because thetraveltime between the aerators is usually a fraction ofthetotaltraveltime(10-30minutes),bacteriahaveonlyafewminutesto shift from aerobic to anoxic conditions and vice versa. For this reason, researchers like Applegate et ah, (1980) and Rittman and Langeland (198S) argue that the most likely explanation of the simultaneous nitrification-denitrification processes taking place in oxidation ditches is that denitrification occurs continuously in the anoxic microzones within thebiological floe. In general, duetothe efficient nitrification and denitrification processes, which simultaneously take place in oxidation ditches, 77V in the effluent is expectedtobeaslowas3mg/1 (OrhonandArtan, 1994). Secondly, oxidation ditches produce lessexcess sludge.Becausethey usually workatan extended aeration mode (i.e.high sludgeresidencetimeand lowfood to micro-organism ratio),oxidation ditchesyield awell-stabilised sludgethathas littleodourproblems (Van der Geest and Witvoet, 1977). This sludge is usually ready for land application. If necessary, however, various chemical treatment or storage with or without dewatering can be used for reducing significantly the amountofpathogenic organisms prior to land applications (Novak et a!., 1984). Note that removal of toxic compounds and heavy metalsisusuallymoreimportantthanreductionofpathogenicorganismsorvectors. Thirdly, due to the high internal recirculation rate, oxidation ditches have good mixing andgoodbuffer against shockloads.Wastewater isusually circulated aroundtheditchin 10to 30minutes (Rittman andLangeland, 1985).The exacttimeneededfor completing one cycle depends on the number of aerators and the dimensions of the ditch. High internalrecirculation coupled withthehighturbulence induced neartheaerators resultin agoodmixingoftheditchcontents. Fourthly; construction of oxidation ditches usually is relatively cheap. The size of the oxidation ditch is usually less than the size of an up-graded multi-stage conventional WWTP that can achieve the same degree of nutrient removal as oxidation ditches. UpgradingofaconventionalWWTPisusuallyachieved byaddition ofatleasttwomore reactors (anoxic and anaerobic reactors) in pre- or post denitrifcation modes. Furthermore, unlike post-denitrifcation systems, oxidation ditches do not need external C-source, as the influent wastewater will be used as a C-source. This simply means additionalsavinginthecapitalcostsbynotinstallingC-sourcefeeding equipments. Finally, operation of oxidation ditches costs relatively less than the operation of other conventional WWTP's. Thehigh rateofnitraterecirculation inoxidation ditchesusually leadstoa significant reduction intheamountofoxygen needed for oxidation, asnitrate,
instead ofoxygen,isused asaterminalelectronacceptor inthedenitrification processes. Furthermore, manpower requirements are minimal and limited to usual cleanings, maintenance,andmonitoringprocedures(Petersenetal., 1993). 1.1.3 Needfor advanced controlin WWTP's Thegrowing interest in the useofadvanced control techniques in biological wastewater plants is mainly motivated by the process complexity and the strict effluent standards (Lindberg, 1997; Andrews 1998; Olsson and Newell, 1998;Lukasse, 1999). Activated sludge processes are quite complex. Many factors affect the performance of activated sludge systems. Examples of these factors are: organic and inorganic loading, sludge viability,oxygenuptakerate,mixing,detentiontime,sludgesettlingpropertiesandsolids level intheclarifier. Optimumperformance ofthesesystems usuallyrequires monitoring and manipulating of certain process variables such as:F/M ratio, oxygen input, recycled activated sludge (RAS) flow, waste activated sludge (WAS) flow, and sludge blanket depth of the secondary settler (Eckenfelder et al., 1986). Biological nutrient removing plants, like oxidation ditches, have even more complex processes. Nitrogen removal processes (nitrification and denitrification) are sensitive to many process and environmental variables such as:DO,pH, temperature andthepresence of inhibitors, hi order to optimize the performance of these complex processes, and to achieve the strict effluent standards,therefore,theuseofadvancecontroltechniquescanbebeneficial. TheuseofadvancedcontrolsystemsinWWTP'shasevenmorebenefits thantooptimise theprocessandtohelptoachievethestandards.Forexample,italsohelpstoincreasethe amount of wastewater processed per unit capacity, and to minimise the number of operating personnel and to increase their productivity. Although the use of advanced control systems in WWTP's has significant benefits, it is constrained by the following main factors: (i) most of the plant operators do not have adequate training in instrumentation and control, (ii) there is a communication problem between the environmental engineers and control engineers, (iii) there is a lack of reliable on-line sensors and (iv)there is a lack of experimental proof oftheproposed control strategies. For more information about the benefits and constraints of the use of advanced control techniques in WWTP's see for example Olsson and Newell (1998). Development of benchmarks, such as inthis thesis,helps to alleviate someofthe constraints that hinder theapplicationofadvancedcontrolsystemsinWWTP's. Literature reviews carried outbyLindberg (1997) and Weijers (2000) show that control strategies proposed for use in oxidation ditch plants arethe same as those proposed for other activated sludge systems. This simply means that there is no control strategy that addresses the special features of oxidation ditches, such as the effect ofthe coupling of oxygen input and horizontal velocity (flow recirculation) on the nitrogen removal processes.Thisthesistriestoaddresssomeoftheparticularfeatures ofoxidationditches. Insection 8.2 ofthisthesis,theeffect ofthehorizontalvelocity,which isconsidered asa controlvariable,isstudiedfornitrogenremovalprocesses.
12 Defining thebenchmarking problem Due to the increased public awareness of the problem of water pollution, effluent standards for WWTP are becoming more and more stringent (EC, 1999;UNEP, 1999). Thistrend isexpectedto continue inthefuture (Olsson andNewell, 1998).Asargued in the previous section, to achieve these strict standards, at minimum costs, advanced control isnecessary. Therefore, numerous control strategies havebeen recently proposed (Lindberg, 1997;Lukasse, 1999;Singman, 1999;Weijers, 2000).However, few ofthese control strategieshavebeen thoroughly evaluated, either inpracticaltests orin computer simulations(Alexetal, 1999;Ponsetal, 1999). Comprehensive evaluation of proposed control strategies is obviously not a trivial task. Due to time and money limitations, evaluation of all the proposed control strategies by carryingoutpracticaltestsisclearlyimpossible.Thuscomputersimulationsoffer a useful approach to solve this problem. However, this approach requires development of a standard simulation procedure in conjunction with standard evaluation criteria. That is, development of a whole benchmarking methodology that can be used in evaluating all proposed control strategies. In this direction, both the European Concerted Action Programme (COST)624 and the IWA Working Task on Respirometry have proposed benchmarking as a tool to evaluate the performance of activated sludge WWTP's (Keesmanetal, 1997;Ponsetal, 1999;Copp,2000). Thetermbenchmarkisfrequentlyusedincivilengineering,particularly insurveying,and also incomputertechnology. Insurveying,abenchmark isapointwithaknownreduced level (height relative to sea water surface), relative to which levels of other points will then be measured. In computer technology, a benchmark is a reference performance to which the relative performance of hardware or software can be assigned. A dictionary definition ofbenchmarkis"Areferencevalueagainstwhichameasurementoraseriesof measurementsmay be compared" Parker (1994). COST624 defines the benchmark as "protocolto obtaina measureofperformanceof controlstrategiesfor activatedsludge plants basedon numerical,realisticsimulationsof thecontrolledplant'. According to thislastmentioned definition, thebenchmark consistsofa description oftheplantlayout, asimulation modelanddefinitions of(controller)performance criteria.
13Research objectives 13.1General objective The main goal of this research is to develop a benchmarking methodology that can be used in evaluating existing or new control strategies proposed for full-scale oxidation ditchWWTP's.
13.2 Specific objectives (i) (ii) (iii) (iv) (v) (vi)
Develop, onbasis oftheavailablerealprocess data,a simple, acceptableand realistic model that adequately describes both the biochemical processes and thehydraulics,andissuitableforcontrollerdesign, Developperformance evaluationcriteriaforoxidationditches, Conduct sensitivity analysis to specify model parameters that need special attentionfromthebenchmarkuser, Carry out uncertainty analysis to quantify the possible effect of the various uncertainty sourcesontheperformance indices, Testtheapplicabilityofthebenchmark, Illustrate the implementation of the benchmark, by evaluating a number of controlstrategies.
1.4Contributionofthethesis By developingthebenchmarking methodology, this thesis will contribute to (i) bridging the gap that exists between control theory and its application in the field of wastewater treatment, (ii)promoting theacceptance ofexisting control strategies and (iii) enhancing thedevelopmentofnewcontrolstrategies.Inparticular, carryingoutitems(ii)and(iii)of the specific objectives, mentioned above, will constitute a major innovation over the known benchmarkingprocedures(seefor exampleCOST(2000)). 1.5Research methodology As mentioned before, the objective of the research is to develop a methodology for evaluating control strategiesused inoxidation ditchWWTP's. Thisobjective isachieved by outlining the procedure for developing such a benchmark for a specific full-scale WWTP,usingtheavailableoperationaldata.ForotherWWTP's,thesameprocedurecan befollowed. Theoxidationditchplant,fromwhichtheoperationaldataused inthisthesis were obtained, is a 300 000 p.e. carrousel WWTP situated in Rotterdam, The Netherlands. Description of this treatment plant, with its existing control strategy, is given in the Appendix IV. It must be emphasized that this plant will notbe a reference case (benchmark). Rather, the procedure developed here can be used in benchmarking anyspecific oxidation ditchWWTP. Thebenchmarking methodology followed here is different from thatproposed by COST 624 and theIWATask Group onRespirometry. Thesegroupshaveproposed to develop benchmarks for hypothetical WWTP's wim typical design capacities. Further, they use typical influent and operational data for developing these benchmarks. In contrast, throughout this thesis real data have been used. Thus, the benchmarking approach followed here is more realistic and suitable for adaptation to other real WWTP's. Adjustments for other oxidation ditches is mainly regarding the aeration and the
hydraulics. In chapter 2, a method for modeling the aeration, using a loop-of-GSTR's model, in terms of the aeration constant (k =KLa-VA) is presented. Here VA is the effectively aerated volume around an aerator. In chapter 3, the adequate number of CSTR'sneededformodelinganoxidationditchisinvestigated. 1.6Researchfocusandlimitations Theresearch isfocused onbenchmarkingcontrolstrategiesused inoxidationditchplants thatperform onlycarbonoxidationandnitrogenremoval.Therefore, phosphorus removal is considered tobebeyondthescopeofthis research.Forthisreason,thefirstversionof activated sludge models,ASM No. 1(Henze et al, 1987), is considered to be sufficient for modeling biochemical processes taking place in the ditch. In addition, the study is limited to oxidation ditches that treat mainly domestic wastewater. Therefore, oxidation ditches used for other purposes, e.g. for treating industrial wastewater, were not studied. The secondary settler was modelled as a non-reactive settler, using the 10-layer one dimensional settler model withthe double exponential settlingvelocity function (Takacs etal, 1991).Sotheemphasisofthisstudywasonthebiologicalprocessestakingplacein theaerationtank.
1.7Outlinesofthethesis Inthisthesis,eachchaptercanberead independently, becausetheyarepresentedasthey havebeen(orwillbe)published. The chapters are grouped intothreeparts. Thefirst part (chapters 2,3 and 4) deals with the modeling issue of oxidation ditch plants. The loop-of-GSTR'smodel, without back flows, was chosen for modeling oxidation ditches because it is simple,realistic,and can be easily incorporated within control algorithms. Chapter 2 investigates the use of the loop-of-GSTK'smodel for modeling oxidation ditches under clean water conditions,and estimates the ditch hydraulics and aeration. Chapter 3 studies the effect of number of CSTR'sonmodelingoxidationditchesbased on influent-effluent concentrations. Chapter 4 presents a new calibration methodology, which can be used in calibrating non-linear systems like the oxidation ditch systems. The methodology that is based onelliptical analysis of response surfaces, is used in calibrating a loop-of-CS7K'.s model used for modeling a real full-scale oxidation ditch plant,under process conditions. According to the existing calibration methods (see for example STOWA (2000)), which assume mat Km is known in advance, first sludge production will be calibrated, then ammonia and finally nitrate.In contrast,thenewmethod allowsa simultaneous calibration ofthethree above-mentioned functions plustheaeration. The second part of the thesis (chapters 5 and 6) analyses the performance of the developed loop-of-CSTi?'smodel. In this part, sensitivity and uncertainty analysis were carried out to assess the reliability and applicability of the developed model, using ellipticalanalysis. Chapter 5 deals with the sensitivity analysis. That is, assessing the
effect of parameter variations on the performance indices. Chapter 6 is devoted to uncertainty analysis. In this chapter, the effect of various uncertainty sources on the performance indices is quantified (see section 6.1). In chapter 6, also a novel backward uncertainty propagation method is illustrated with a working example (see section 6.2). Backward uncertainty propagation gives essential information for reduction of the predefined parameteruncertainty regionandparametersdominatingspecific phenomena. Thethirdpartofthethesisprovidesthebenchmarkingprocedureandillustratestheuseof the benchmark. In chapter 7, components ofthebenchmark are defined and the step-bystep benchmarking procedure is outlined. In chapter 8, the use of the benchmark is illustrated.Insection8.2,thebenchmarkisusedfor evaluatingtheeffect ofthehorizontal velocity on the performance of oxidation ditches. Here, the horizontal velocity is considered asa control variable,from73Vremoval efficiency pointofview. In oxidation ditches,oxygen inputand flow recirculation (horizontal velocity)are coupled,duetothe use of mechanical aerators. At high horizontal velocity, high amounts of nitrate and dissolved oxygen will be recirculated from aerobic zones to the anoxic zones. Consequently, nitrogen removal processes will significantly deteriorate. In section 8.3, the usefulness of the benchmark is illustrated by using it to evaluate some basic and advanced control strategies. Finally, chapter 9 ends the thesis with overall discussion, conclusionsandrecommendationsforfuture research.
1.8 References Alex,J.,J-.F.Beteau,J.B.Copp,C.Hellinga,U.Jeppsson, S.Marsili-Libelli,M.N.Pons, H. Spanjers, and H. Vanhooren (1999), Benchmark for evaluating control strategies in wastewater treatment plants, ECC'99 (European control conference), Karlsruhe August 31-September3,1999. Araki,H.,K. Koga,K. Inomae,T. Kusuda,and Y. Awaya (1990),Intermittent aeration fornitrogenremovalinsmalloxidationditches,Wat.Sci.Tech.22(3-4):131-138. Andrews, J.F. (1998), Dynamics and control of wastewater treatment systems: an overview, in dynamics and control ofwasewater systems,Water Quality library, Vol. 6, 21"1 edition, edited by: M.W. Barnett, M.K. Stenstrom, and J.F. Andrews, Technomic PublishingCompany,Inc. Applegate,C.S.,B.Wilder,andJ.R.DeShaw(1980).Totalnitrogen removal inamultichanneloxidationsystem,JWPCF52:568-579 CBS (2000), Waterkwaliteitsbeheer: zuivering van afvalwater, 1998, Centraal Bureau voorStatistiek,Voorburg/Heerlen(inDutch). Copp, J.B. (2000), Defining a simulation benchmark for control strategies, Water21, l(l):44-49.
COST(2000),www.ensic.u-nancy.fr/COSTWWTP/Benchmark/Benchmarkl.htm. DHV Water (1993), dossier G8112 - 10-001, DHV Water BV, Amersfoort, The Netherlands Drews,R.J.L.C.,W.M.Malan,P.G.J.Meiring,andB.Moffatt (1972),Theorbalextended aerationactivatedsludgeplant,J.WPCF7:1183-1193. EC (1999), Implementation of Council Directive 91/271/EEC of 21 May 1991 concerningurbanwastewatertreatment,asamended byCommission Directive98/1S/EC of 27 February 1998:summary ofthe measures implemented by the member states and assessment of the information received pursuant to article 17 and 13 of the Directive, European Commission, Office for Official Publications of the European Communities, Luxembourg. Eckenfelder, WW., D.L. Ford, P.W. Landlord, G. Shell, and D.L. Sullivan (1986), Operation, control,and management ofactivated sludgeplants,Departmentof Civil and EnvironmentalEngineering,VanderbiltUniversity,Nashville,Tennessee,USA. Fair,G.M.andJ.C. Geyer,(1958),Elements ofwatersupplyand wastewaterdisposal,5th ed,JohnWiley&Sons,Inc. Henze, M., C.P.L. Grady, jr, W. Gujer, G. van R Marais and T. Matsuo (1987), Activated sludge model no. 1, IAWQ scientific and Technical Report no. 1, JAWQ, London,U.K. Huang, J.Y.C. and D.M Drew (1985), Investigation of the removal of nitrogen in an oxidationditch,J.WPCF57(2):151-156. Inomae,K., H. Araki,K. Koga, Y. Awaya, T.Kusuda, and Y.Matsuo (1987),Nitrogen removalinoxidationditchwithintermittentaeration,Wat Sci.Tech. 19:209-218. Keesman, K.J., H. Spanjers, P. Vanrolleghem, and J. Alex (1997), Benchmarks for control strategies in activated sludge plants, TMR Research Proposal No. ERB4061PL970020,Wageningen,TheNetherlands. Koot, A.C.J. and J. Zepers, (1972), Carrousel, a new type of aeration-system with low organicload,Wat Res.6:401-406. Lester,J.N.(1996),sewageandsewagesludgetreatment,inPollution:causes,effects and control, 3rd ed., edited by RM. Harrison, the royal society of chemistry information services. Lindberg, C-F. (1997), Control and estimation strategies applied to the activated sludge process,Ph.D.thesis,UppsalaUniversity,Sweden.
10
Lukasse, L.J.S. (1999), Control and identification in activated sludge processes, Ph.D. thesis,WageningenUniversity,TheNetherlands. Metcalf & Eddy (1991),Wastewater engineering: treatment, disposal and reuse, 3rd ed, McGraw-Hill. Novak,J.T.,M.P.Eichelberger, S.K. Banerij,andJ.Yaun(1984),Stabilisation of sludge from anoxidationditch,J.WPCF56(8):950-954. Olsson, G. and B. Newell, (1998),Wastewater treatment systems: modeling, diagnosis, and control,IWAPublishing,London,UK. Orhon, D. and N. Artan (1994), Modeling of activated sludge systems, Technomic PublishingCompany,Inc.Lancaster,Pennsylvania,USA. Parker,S.P.(1994),McGraw-HillDictionaryofScientific&TechnicalTerms,5thedition. Pasveer,A (1971),Verdereontwikkehng.Hetoxydenitroproces,H 2 0 4(22):499-504,(in Dutch). Petersen, G., H. Nour El-Din,and E.Bundgaard (1993), Second generation of oxidation ditches:advancedtechnology insimpledesign,Wat.Sci.Tech.27(9):105-113. Pons, M.N., H. Spanjers and U. Jeppsson, Towards a benchmark for evaluating control strategies inwastewater treatmentplants bysimulations,Escape9(European symposium oncomputeraidedprocessengineering-9),Budapest(1999). Qasim, S.R.(1999),Wastewatertreatment plants:planning,designandoperation, 2nded, TechnomicPublishingCO.,Inc. Rittman,B.E.andW.E.Langeland (1985),Simultaneous denitrification with nitrification insingle-channeloxidationditches,J.WPCF57(4):300-308. Takacs, I , G.G. Patty and D. Nolasco (1991). A dynamic model of the clarificationthickeningprocess,Wat.Res. 25(10):1263-71. Sen, D., C.W. Randall, T.J. Grizzard and D.R. Rumke (1992), Process design and operation modifications of oxidation ditches for biological nutrient removal, Wat. Sci. Tech.25(4-5):249-256. Singman,J.(1999),Efficient controlofwastewatertreatmentplants- abenchmarkstudy, M.Sc.thesis,UppsalaUniversity, Sweden. STOWA (2000), SJMBA-protocol, richtlijnen voor het dynamisch modelleren van actiefclibsystemen, STOWA,2000-16(inDutch).
11
Van der Geest, AT. and W.C. Witvoet, (1977), Nitrification and denitrification in a cairouselsystem,Prog.Wat Tech. 8(4/5):654-660. UNEP (1999), Global environment outlook 2000, United Nations Environment Programme,EarthscanPublicationLtd,London. WEF (1998), Biological and chemical systems for nutrient removal, A special publication, Municipal subcommittee of the technical practice committee, Water EnvironmentFederation(WEF),USA. Weijers, S. (2000), Modeling, identification and control of activated sludge plants for nitrogenremoval,Ph.D.thesisEindhovenUniversityofTechnology,TheNetherlands.
12
PARTI MODELING
13
2.Oxygentransferrateestimationinoxidationditchesfrom cleanwatermeasurements1 2.1Abstract Standard methodsforthedetermination ofoxygentransfer ratearebased onassumptions that are not valid for oxidation ditches. This paper presents a realistic and simple new method to be used in the estimation of oxygen transfer rate in oxidation ditches from clean watermeasurements.Thenewmethodusesaloop-of-C£ZR's model,which canbe easily incorporated within control algorithms, for modeling oxidation ditches. Further, this method assumes zero oxygen transfer rates (Ki/x) in the unaerated CSTR's. Application of a formal estimation procedure to real data revealed that the aeration constant (KLaVA, where VAis the volume of the aerated CSTR), can be determined significantly more accurately than Kifl and VA. Therefore, the new method estimates k instead ofKjja. From application to real data, this method proved to be more accurate thanthecommonlyusedDutch standardmethod(STORA,1980). Keywords;oxygentransfer rate,aeration,Ki^,oxidationditch,carrousel. Nomenclature C,:DO concentrationinoutflowofther*CSTR (mg/l). C,.,:DO concentrationintheinflowofthethCSTR (mg/l). COD: chemicaloxygendemand(mg/l). Cs:DO saturationconcentration (mg/l). CSTR: completelystirred tankreactor. DO: dissolvedoxygen. J:objectivefunction. k aerationconstant,KLa •VA, (m3/min) kjo:aerationconstantat 10"C(m'/min). Kifl-.overalloxygentransferrate(miri) . NotethathereK.uaiscalculatedfor theassumedaeratedvolume andnotfor thewholeditch, m:numberofaerators. N:numberofobservationsamplesinstances, n:numberofcompartments(CSTR s). OCio: aeratoroxygenation capacityat 10°C. OTR:oxygen transferrate(kgOJh). q:waterflow (m'/min.). rpm:revolutionperminute. SOTR:standardoxygentransferrate(kgOJh). t:time (min.). VA:volumeoftheaeratedCSTR (m3). V„tx.A mixedvolumearoundtheaeratorfin3). VnA: volumeofthenon-aeratedCSTR (m ) . VTOT'- totalvolumeoftheoxidationditch (m'). f
AslightlymodifiedversionpublishedbyA.Abusam,K.J.Keesman,K.MeinemaandO.vanStratenin Wat.Res.35(8)2058-2064.
14
V. = Vjoi/n
X:sensitivitymatrix. y: measuredDO concentration (mg/l). v indexthatindicateswhethertheCSTR isaerated (v=l) or unaerated (r=0). ftparameter vector cov ft covariancematrixof6. £: residuali.e.measuredDO-estimated DO(mg/l).
*
150
200
250
300
measured predicted
350
400
400
400
Fig.8,Realeffect ofseasonalchangesinwatertemperatureontheperformance oftheoxidationditch
Fig. 9 presents the results obtained from one simulation of the oxidation ditch, using nominal values (Table 1), Eqn. (7) and (8) and average daily influent data. From comparison withFig. 8,this figure clearlydemonstrates thatthemodel is abletopredict approximatelytherealseasonalpatternsfor TNandCODremovals.
70
Thus it can be concluded that uncertainty due to seasonal changes in temperature can be quantified, inthe sameway as for the other uncertainty sources illustrated above, by Monte Carlo simulations. 150
200
250
For illustrating how Urn* (day) uncertainty due to seasonal Fig.9,Predictedeffect of seasonaltemperature changesonthe performance oftheoxidationditch changes in temperature can bequantified, the following assumptions were first made: (i)the change inwatertemperature is between 7to 30°C, and (ii) temperature activity coefficient (0) varies between 1 to 1.08, for kinetic parameters,and between 1 to 1.047,fortheaeration constant(k).Here itisworthtonote thefollowing. First,Fig.7showsthatin 1993,temperaturevariedbetweenabout 10to22 °C,buthereweareusingtemperaturerangeof7to30°Ctostudythepossibleeffect that may take place during a number of years. Second, here we are not investigating the temperature effects onaparticularday,rather weare estimatingtheeffects overaperiod ofseven days,which can be inthesummerorwinter season,under constant temperature conditions. Then, as before, the LHS technique was used for generating 500 uniformly distributed samples from the assumed ranges. Finally, Monte Carlo simulations were carriedout,usingtherealdatascenario. 6.1.5.4.2Resultsanddiscussion Results of simulations are presented in Fig. 10. As for theother uncertainty sources, the effect of uncertainty due to seasonal changes in water temperature is greater on the performance indices EQ and TSPthanontheother indices (see e.g. C.V., coefficient of variation, given in Table 2). However, unlike other uncertainty sources, uncertainty in water seasonal temperature has induced the highest variation inthe index AE. Of course, this is due to allowing the aeration
mm EflluentQuality(EQ)
1001
'
•
0 5000 10000 15000 Aeration Energy(AE) kWh/d
kcj/a^
•
1
-1000 0 1000 2000 3000 TotalSludge Production(TSP) kg/d
801
•
•
1
TotalDisposable Sludge(TDS)
Fig. 10,Frequencydistributionsoftheperformance indices due to seasonal changes in water temperature between 7 to 30 °C, usingtherealdata scenario.
71
kg/d
constant(k)tovarywiththechangesinwatertemperature. Given the possible ranges of water temperature and possible values of temperature activity coefficients (0's), propagation of seasonal temperature uncertainty can also be quantified togetherwithparametervaluesandinfluentloadsuncertaintiesgiveninsection 4.1. 6.1.6 Generaldiscussion It shouldbenotedthatresultsobtained inthis studyrepresentonlytheshort-term effects of the various uncertainty sources on the performance indices, since the evaluation is carried out for a period of only seven days. Because of the relatively slow process of biomass growth, different results mightbe obtained for the long-term effects. However, thesameproceduresfollowed herecanbeappliedfor studyingthelong-term effects. Note also in this paper, the effects of the different sources of uncertainty on the performance indiceswereindividually quantified. Infact,thisisdoneonlyfor illustration purposes.However, inreality,thebenchmarkuserneedstoquantify thecombined effect ofthe major sources.For achieving that,the user is advised to evaluate the effect ofall thesemajorsourcesatonetime. Estimation ofthe individualuncertainty contributions ofthevarious sources is especially important when one of the sources become actually known. It is also equally important for designing experimental or monitoring programmes with the aim of reducing the uncertainty. In a previous work (Abusametal., 2001),wehave shown thatthe effect of some of the parameters of theASM No. 1 on the performance indices depends on the value of some other parameters. Thus for computing the individual uncertainty contribution ofthevarious sources,thebenchmark user isadvised to used amethod that deals with correlation cases,e.g. the so-called "partial uncertainty contribution method' (Johansson and Janssen, 1994). For more about uncertainty reduction, the benchmark user is referred to e.g. Van Straten and Keesman (1991)whodescribe a full strategy for reducinguncertaintyin forecasting. For the benchmark user, practical implications of accurate estimation of uncertainty propagation are the following. First, selection can be made, under uncertainty, among various control strategies. Secondly, decision can be made about the usefulness of a certaincontrolstrategy,whichisclaimedtobeuseful undercertainconditions. 6.1.7 Conclusions This study has demonstrated how the effect of the various uncertainty sources can be quantified. Uncertainty sources considered are: (i) parameter values, (ii) influent loads, (iii) values of the initial states, (iv) model structure, and (v) seasonal changes in water
72
temperature. Although onlytheshort-term effects werestudiedhere,thesameprocedures canalsobeappliedforstudyingthelong-term effects. Short-term results obtained have indicated the following. Due to uncertainty in influent loads andparameter values,large deviations,from thenominal values,inthe benchmark performance indiceswillbefound for effluent qualityandtotalsludgeproductionindices. However, relatively smaller deviations will be found due to uncertainty in the states initial conditions. Effect of the model structural uncertainty on the performance indices seemstobenegligible.
6.1.8 References Abasaeed,E.A.(1997),SensitivityanalysisonasequencingbatchreactormodelI: effect ofkineticparameters,J.Chem.Tech.Biotechnol. 70:379-383. Abasaeed, E. A. (1999), Sensitivity analysis on a sequencing batch reactor model II: effect of stoichiometric and operating parameters, J. Chem. Tech. Biotechnol. 74:451455. Abusam,A.,K.J.Keesman,G.vanStratenandK.Meinema(2000),Parameter estimation procedure for complexnon-linear systems:calibration of ASMNo.l for N-removalina full-scale oxidationditch,inProc.oftheWatermatex2000,Gent,Belgium. Abusam, A.,K.J. Keesman, G. van Straten and K Meinema (2001), Sensitivity analysis on oxidation ditches: the effect of variations in stoichiometric, kinetic and operating parametersontheperformance indices,J.Chem.Tech.Biotech.76(4):430-438. Bagchi,A.(1993),Optimal controlofstochasticsystems,PrenticeHall. Beck, M.B. (1987), Water quality modeling: a review of the analysis of uncertainty, WaterResources.Research23(8):1393-1442. Burges, S.J. and D.P. Lettenmaier (1975), Probalistic methods in stream quality management,Wat.Res.Bulletin11:115-130. Copp, J.B. (2000), Defining a simulation benchmark for control strategies, Water21, l(2):44-49. DHVWaterBV(1993),DossierG8112-10-001,Amersfoort,TheNetherlands. Henze,M , C.P.L. Grady,jr,W.Gujer, GvanR.MaraisandT.Matsuo (1987),ageneral modelforsingle-sludgeWWTsystems,WatRes21(5):505-15.
73
Haybye,J.A. (1998),Model error propagation and data collection design:an application inwaterqualitymodeling,Water,AirandSoilpollution 103:101-119. Janssen P.H.M. (1994), Assessing sensitivities and uncertainties in models: a critical evaluation, in predictability and non-linear modeling innatural sciences and economics, EdbyGrasmanJandvanStratenG.,pp344-361,Kluwer AcademicPublishers. Johansson, M.P. and P.H.M. Janssen (1994), Uncertainty analysis on critical loads for forest soils inFinland, inpredictability andnon-linear modeling in natural sciences and economics, ed by Grasman J and van Straten G., pp 447-459, Kluwer Academic Publishers(1994). Metcalf &Eddy (1991), Wastewater engineering: treatment, disposal and reuse, 3 rd ed, McGraw-Hill. Pons M. N.,H. Spanjers and U. Jeppsson, Towards a benchmark for evaluating control strategies inwastewater treatmentplants bysimulations,Escape9(European symposium oncomputeraidedprocessengineering-9),Budapest(1999). Spanjers, H., P. Vanrolleghem, K. Nguyen, H. Vanhooren and G.G. Patry (1998), Towards a simulation-benchmark for evaluating respirometry-based control strategies, Wat.Sci.Tech.37(12):219-226. Vanrolleghem, P.A., U. Jeppsson, J. Carstensen, B. Carlsson and G. Olsson (1996), Integration ofWWTPdesign andoperation- asystematicapproach usingcost functions, Wat.Sci.Tech.34(3-4), 159-171(1996). Van Straten, G. and K.J. Keesman (1991), Uncertainty propagation and speculation in projective forecasts of environmental change: a lake-eutrophication example, J. of Forecasting 10:163-190. Von Sperling,M.(1996),Designoffacultative pondsbased onuncertaintyanalysis,Wat Sci.Tech.33(7):41-47. Weijers, S.R.andP.A.Vanrolleghem (1997),Aprocedureforselectingbest identifiable parameters in calibrating activated sludge modelNo. 1to fall-sale plant data, Wat. Sci. Tech.36(5):69-79.
74
6.2Forwardandbackwarduncertaintypropagationin mathematicalmodels** 6.2.1Abstract In the field of water technology, forward uncertainty propagation is frequently used, whereas backward uncertainty propagation israrelyused.In forward uncertaintyanalysis, one moves from a given (or assumed) parameter subspace towards the corresponding distribution of the output or objective function. However, in the backward uncertainty propagation,onemovesinthereversedirection,fromthedistribution function towardsthe parameter subspace. Backward uncertainty propagation, which is a generalisation of parameterestimation erroranalysis,givesinformation essentialfor designingexperimental or monitoring programmes, and for tighter bounding of parameter uncertainty intervals. The procedure of carrying out backward uncertainty propagation is illustrated in this technicalnotebyaworkingexample. Keywords;wastewater;activatedsludge;oxidationditch;modeling;uncertaintyanalysis. 6.2.2 Introduction Uncertainty analysis isavery important step inthe modelbuilding process.It contributes directly to the reliability and applicability of the developed mathematical model. It is mainly concerned with the effect that various sources of uncertainty have on the model output Model sources of uncertainty can be in: (i) model inputs, (ii) model parameter values, (iii) initial state conditions,and in(iv)model structure. The method illustrated in thispaperisparticularlyapplicabletotypes(i)to(iii)ofsourcesofuncertainty. Inthefield ofwatertechnology, uncertainty analysis orerrorpropagation- ifit iscarried out at all - is usually executed in one direction: forward direction. That is, starting the uncertainty analysis from a given (or assumed) parameter subspace, defined in terms of rangesordistributions,andmovingtowardsthecorrespondingdistributionoftheoutputor objective function. However, the backward uncertainty propagation is rarely performed. Clearly the backward uncertainty propagation is the reverse of the forward uncertainty propagation, and it can be seen as a generalisation of parameter estimation error quantification from given experimental data.It can beused for obtaining, in a systematic way, essential information about which part of the parameter space, or which parameter combinations, contributed mostly to some interestingpartofthedistribution function. For instance, which inputs, parameters or initial conditions lead to extreme or off-normal process conditions? Such information willbethe important ingredients thathelptheplant manager in designing and carrying out a monitoring programme. Through a monitoring programme, the plant manager usually wants to find out accurate values for model parameters or initial conditions that are suspected of causing or contributing to a certain part of interest in the distribution function found from forward uncertainty propagation. ** SubmittedtoWat.Res.byA.Abusam,K.J. KeesmanandG.van Straten
75
This can even further generalized by formulating extreme or off-normal process conditions,apriori. 6.2.3 Theory Letthemodelofthedynamicsystembedefined asasetofequationsinstandardstatespaceform: dx
^ =f[x,u,t;0] +w(t), x(0)=x0 (1) at y(t) =g[x,u,t;0] +v(.t) (2) where both/ a n d g are vector-valued functions, x isn-dimension state vector with initial statevectorxo, uisthew-dimensioninputvector, 0isthe^-dimensionparametervector,y isthe^-dimensionoutputvectorandvandwarestochasticsignalsrepresentingthesystem noise and the measurement errors, respectively. Further, let there be an interest in the performance oftheplant,expressedbysomeobjectivefunction definedas: T
J ={y,u,t;0)+JL[y, u,t;0}dt
(3)
o
where definestheterminalcost,andL defines therunningcost Noticethataccordingto (3)J is a real-valued scalar function. Extension to a vector-valued objective function is straightforward, asinAbusam(2000). Parameter space, P
JL
"prior
m
^
1 i
JpotUrJor
•
'
1
.
-
\ I
I
I
