Integration of rulebased expert system casebased easoner and an ontological kno wledgebase - PDF document

Integrationofarule-basedexpertsystem,acase-basedreasonerandanontologicalknowledge-baseinthewastewaterdomainLuigiCeccaroniAbstract.Wepresentanenvironmentaldecision-supportsystemintegratingarule-basedexpertsystem,acase-basedreasonerandanontologicalknowledge-base.Thissystemisabletomodeltheinformationaboutawastewatertreatmentprocessthroughthedef-initionofthebasictermsandrelationscomprisingthevocabularyofthewastewatertreatmentarea.Furthermore,thismanagementsystemoptimizestheoperationofwastewatertreatmentbyamorereliablemanagementandmakingeasieritsself-portability.1INTRODUCTIONThegeneralissueswewouldliketoaddressinthepaperare:*optimizingwastewatertreatmentoperationbyamorereliablemanagementand*makingeasiertheportabilityofthemanagementsystem.1.1PhysicalenvironmentWastewaterpuricationContaminationlevelsofwaterscon-stantlyincreaseduebasicallytoindustrialdevelopmentandtotheincreaseofpopulationdensityincertainzones.Wastewaters,eitherindustrialorurban,havetobedecontaminateduntilanadequatelevel,sothattheycouldbepouredtothesurroundinghydricmediumwithoutcausingproblemsofenvironmentaldeterioration.Forthat,therearewastewatertreatmentplants(WWTPs)withtechniquesofphysical-chemicalandbiologicaltreatment.Everytreatmentprocesscarries,ingreaterorminormeasure,eco-nomicandenvironmentalcostsasforthatitgeneratesanotherkindofwastethatneeds,inturn,ofothereliminationtechniques.Inthissense,forthecaseofwastewaters,theuseofbiologicalprocessesofpuricationisfavoredoverthephysical-chemicalones.Biologi-calprocesses,ingeneral,almostdonotconsumereagents,theyaremoreefcient,theydonotgenerateeithergasesornoxioussludges,andtheyareresponsibleofahigherproductionofsludgesthatcanbeused(notalways)inotherproductiveprocesses(e.g.asfuel,fertilizerandllingmaterial).Wastewatertreatmentplantswithbiologicalprocessesarethephysicalenvironmentmodeledbyoursystem.ThegeneraloperationofaWWTPalwaysincludesvariousinternalpre-designedstandardunits,whosesub-operationisoptimizedtoaccomplishasingletask2
UniversitatPolitecnicadeCatalunya,DepartamentdeLlenguatgesiSis-temesInformatics,CampusNord,ModulC6,JordiGirona3,08034Barcelona,Spain,email:luigic@lsi.upc.esAtaskinthiscontextisgenerallytheremoval/remediationofcontaminantsubstancesorpathogenicmicroorganisms.[13].Eachsub-operationusuallyhaseffectsonotherdownstreamtreatmentprocesses,andtradeoffsbetweenincreasingtheefciencyofoneprocessoranotherarenecessary,takingintoaccountmajorconstraintssuchaswatercharacteristics,efuentqualityandcostsofeachoperation.1.2SoftwareenvironmentThesystemwepropose(namedDAI-DEPUR+)receiveson-linein-puts3fromsensorsallovertheWWTPaswellasoff-lineinputsfromtheWWTPlaboratoriesandhumanoperators.Thesystemusesitsin-ternalknowledge-bases4andinference5mechanismstoprocessandunderstandthisinformation,todiagnosetheongoingWWTP-state,andtopredicttheevolutionofthatWWTPstate.Eventually,theout-putofthesystemisrepresentedbystatementsaboutactionstobetaken,orstatementstosupporthumandecisionsinfutureactuations,ordirectcontrolsignalstoWWTPdevicesinordertomaintaintheplantworkingcorrectly.Incaseofdiagnosisimpasse,DAI-DEPUR+,beforeturningtotheplantmanager,willtrytosolvetheproblemexploitingtheconnec-tion,intheontology,betweendataandstatesoftheWWTP.1.3MotivationsTheprocessofwastewatertreatmentissocomplexthatitisdif-culttodevelopareliablesupervisorytechnologybasedonlyonachemical-engineeringclassic-controlapproach.NowadaystheuseofArticial-Intelligence(AI)systemsseemstobenecessaryinordertoobtainbetterresultsinwastewatermanagement.Rule-basedexpertsystems(oneofthebroadlyappliedparadigmsofAI)provedabletocopewithsomeknowndifcultiesandtofaceseveralWWTP-domainproblems,eveniftheyarenotthedenitivesolutiontothetreatmentproblemasawhole.Ontheotherhand,large,multifunctional,availableontologieswouldsignicantlyim-provecurrentexpertsystemsandtutoringsystemsbecausetheycon-tainthebroadknowledgeofadomainrequiredtoperformmultipleInput/outputdevicesareanyofvariousdevicesusedtoenterinformationandinstructionsintotheDAI-DEPUR+systemforstorageorprocessingandtodelivertheprocesseddatatoahumanoperatoror,insomecases,amachinecontrolledbythesystem.Suchdevicescomprisesensorsandeffectors.ApparatusofthiskindwithdirectconnectiontoDAI-DEPUR+'scentralprocessingunitissaidtobeon-line;peripheralequipmentworkingindependentlyofitistermedoff-line.Bothstatic(rule-based),dynamic(case-based)andontologicalknowledgebases.Inferenceistheprocessofdrawingconclusionsaboutaparticularparameterofthedomain. tasksandtoexplaindomainknowledgefrommultipleviewpoints.Alotofontologiesarenowadaysbeingbuiltinmanyresearchcentersaroundtheworld,butafewofthemarespecializedinbiologyorecology,letalonewastewatermanagementandWWTPmicrobiol-ogy.Agreatimprovementinaddressingthiskindofproblemscancomefromtheintegrationofdifferentmodelingandreasoningsys-tems,suchasspecicontologies,rule-basedreasoning,case-basedreasoningandreactiveplanning.Thearchitectureweadopt(DAI-DEPUR+)integratestwoknowledge-basedsystemsandanontology,andisexibleenoughtodealwiththecomplexityofthewastewatertreatmentprocess,givenanadequateamountandkindofdata.IntheDAI-DEPUR+architecture,withtheembeddedontologyforthewastewatertreatment,therepresentationofadeeperknowledgeofthedomainispermittedandtheevolutionofWWTP-microorganismcommunitiescanbetakenintoaccount.Inthisway,themanagementofbiologicalproblemsarisinginatreatmentplantismoreeffective.Forthersttime,bytheintegrationofanontologywithtwoknowledge-basedsystems,itwillbepossibletocapture,understandanddescribetheknowledgeaboutthewholephysical,chemicalandmicrobiologicalenvironmentofawastewatertreatmentplant.Thebasictermsandaxiomsoftheontologywillentailamodelofwastewaterdomainandaclassicationofthemicroorganismsaccordingtoknownbiologicaltaxonomy,andwillincludeacom-pletedescriptionofthemicroorganismsthemselves(physicalaspect,abundanceandbehaviorinthetreatmentplant).Therelationsoftheontologywillinclude,besidestheclassichierarchicalafliations,alltheinterestingbindingsamongmicroorganismsandbetweenthemandthestateoftheplant(diagnosispotential).1.4GeneraloverviewInthispaperwepresentadecisionsupportsystemforthesupervisionofwastewatertreatmentplants,whichispartoftheknowledgeandtechnologyneededfortherationalmanagementofwaterresources.Westartbydescribing(insection2)relatedworkontheenvironmental-domainstudyandtheAItechniques(includingon-tologies)relatedtothecreationofenvironmentaldecision-supportsystems.Insection3weexplainhowthedecisionsupportsystem(DAI-DEPUR+)hasbeendesignedandweincludeadescriptionofitslayeredarchitecture.Eventually,insection4thecontributionsofthisworkaresummarizedanddiscussed.2WASTEWATERDOMAINANDAITECHNIQUES2.1WastewatertreatmentprocessInthissectionwedescribethegeneraltreatmentprocess,itspossiblevariations,andawastewaterdescriptionfromaphysical,chemicalandbiologicalpointofview.WastewatertreatmentprocessThewastewatertreatmentpro-cessispartofthewatercycleand,assuch,ithasadirectrelationwithotherwatersystemsorreservoirs.Wastewatertreatmentplants(WWTPs)receivewaterfromtheanthropicsystemofsewers,theysomehowprocessit,andnallytheydeliverthiswatertoanaturalreservoir.Thewastewaterprocessingiswhatwecareabout,butwecannotforgetthetwootherclosestcomponentsoftheglobalwatercyclejustmentioned(sewers,andriverorsea).ItisonthebasisofthequantityandqualityofwatertobetreatedthatWWTPsarebuilt,takingintoaccountthepossibleuctuationsintheinow.Theseuctuationscanbeveryimportantwheretheseweragesystemisnotverydevelopedandthereforeitisnotabletodampdowninowpeakstowardstheplant.Themainobjectivesinwastewater-treatmentresearchare:knowingbettertherelevantcharacteristicsofthewastewater,refrainingthecontaminatedwaterfromreachingthenaturalenvi-ronment.Thefactisthatcontinuouslyincreasingeconomicandculturalpressuresonfreshwaterresources,includingpollutionandexcessiveuse,arecausingthreatswhichareaugmentingcostsandmultiply-ingconictsamongdifferentusersofthisstrategicresource.Thesepressurescanalsoimpairthenaturalregenerativefunctionsoftheecosystemsinthewatercycle.Twoofthemainchallengesintheareaofgeneralwater-managementaretoprotectthewaterbodiesandtoprovidehighqualitywaterinsufcientquantityataffordablecosts.Inordertoachievethesegoals,multidisciplinaryresearch-effortsandactionsarenecessary.TheveryexistenceofWWTPsandtheresearchforimprovingthemgoesinthisdirectionandconstitutesanessen-tialelementforanintegratedsustainablemanagementofwaterre-sources.Theobjectivesofsuchsustainablemanagementaretode-veloptechnologiestopreventandtreatpollutionofwater,topurifywater,touseandre-useitrationally,toenhanceefcienttreatmentofwastewaterandtominimizeenvironmentalimpactsfromwastewatertreatment(includingthepreventionofpotentialhealthhazards).2.1.1GeneralwastewatercharacterizationUrbanwastewatercanbecharacterizedinaccordancewiththepres-enceofdifferentkindsofdumping,suchasdomestic,commer-cialorindustrialones.Anotherimportantfeatureisthepresenceofpathogenicorganisms,whichcanprejudiceapossiblealternativereuseoftreatedwater,suchasirrigation.Therearesubstantiallytwocomponentsinwastewater:humanmetabolicwasteanddiscardedmaterial.Whiletherstcompo-nentisalmostchangelessinnature(asitisdependentonhumanmetabolism),thesecondonedependsonmanyparameters,suchasstandardofliving,localhabitsandcountry.2.1.2PhysicalcomponentsTotalsolidscanbedistinguishedinsuspended(sedimentableornot),colloidalanddissolved,andcontainorganicandinorganicportions.Thesizeofthesolidsthatarepresentinwastewaterinuencesthesedimentation,adsorption,diffusion,masstransferandbiochemicalreactions.Thetemperatureofwastewaterdependsonthetypologyofdumpingandonthepermanencetimeinthesewers.Exceptforsummermonths,itishigherthanenvironmenttemperature,duetothepresenceofwarmwaterdumpingfromkitchensandbathrooms.Theimportanceofwastewatertemperatureisboundtothebiolog-icalactivityofpuricationintreatmentplants.Atmorethan40Cnitricationhaltsandtemperatureshigherthan50Cblockaero-bicdigestion.Temperatureslowerthan15Cinhibittheanaerobicmethanogenicprocess,whileat5Cthenitricantautotrophicorastopsitsactivityandat2Calsotheheterotrophicorabecomein-effective.Wastewatercolorisstrictlycorrelatedtoitsage,itssepticconditionsandtothepresenceofindustrialdumping.Theodorisas-sociatedtoputrescenceanddecompositiondegreeoforganicmatter,andtothepresenceofparticularindustrialwastewater.8-2 2.1.3ChemicalcharacteristicsHere,abriefdescriptionoforganicandinorganicchemicaldescrip-torsofwastewaterisgiven.Ingeneral,organicmatterisrapidlybiodegraded,butpartofitisnotandmoreoveristoxicformanyWWTPmicroorganisms.Toeval-uatethecontentoforganicmatter,thebiochemicaloxygendemand6andthechemicaloxygendemand(COD)aredetermined.ThemajorityoftoxiceffectsonWWTP-microorganisms'growthareattributabletoinorganicmatter,suchasheavymetals,andtoitsinteractionwithotherwastewatermaterials.Thenitrogenfoundinwastewaterisofveprevalentkinds:or-ganicnitrogen(invegetalandanimalproteins),ammoniacalnitro-gen,nitrites,nitratesandelementalgaseousnitrogen.Ammoniacalnitrogenisproducedduringthedecomposition/hydrolysisofor-ganicnitrogenandcancomefromthebacterialreductionofnitritesordirectlyfromindustrialdumping.Themainkindsofphosphorusexistinginwastewaterare:saltsoforthophosphoricacid,polyphosphatesandorganicphosphorus.Inurbanwastewater,ingeneral,allkindsofphosphorusarepresent,while,afterabiologicaltreatment,therearegenerallyonlyor-thophosphates.Sulfurispresentintheformofsulfatesorsuldes.Sulfatescanbereducedtosuldesbysulfate-reducerbacteriainanaerobiccon-ditions.Sultesconstituteaculturemediumforseveralspeciesofaerobicbacteriaabletocreatesulfuricacid,whichcancausecorro-sionproblems.Chlorideshavemetabolichumanorigin(astheyarecontainedinurineinanamountequalto1%)orareduetoindustrial-watercon-tribution.SomeheavymetalsinwastewaterarenecessaryinminimumamountsasmicroelementsforWWTPmicroorganismsandforaquaticlife,buttheyarepoisonousinhighconcentrations.2.1.4BiologicalcomponentsAbasicknowledgeaboutthemostcommonnaturalorganismsthatcanbefoundinwastewaterisalsonecessarytocontrolthetreat-mentprocess.Someoftheseorganismsareessentialforcertainpollution-removaltreatments,suchasactivatedsludge.Themajorityofpathogenicorganismsarepartofhumanintestinalbacterialoraandtheycannotsurviveforalongtimeinwastewater.Ingeneral,mostoftheorganismsofhumanoriginarebanalsaprophyticbacte-ria,thatisorganic-matterdemolishers;theyarenotpathogenicandcanenterbiologicalprocesseswithoutanyproblem[12].2.1.5WastewatertreatmentplantsInawastewatertreatmentplant(WWTP),themaingoalistoreducethelevelofpollutionoftheinowwater,thatistoremove,withincertainlimits(dependingonlocallegislation),abnormalamountsofpollutantsinthewaterpriortoitsdischargetothenaturalenviron-ment.Thiscanbedoneinanumberofdifferentways,correspondingtodifferentkindsofWWTP.ThemostwidespreadclassesofWWTPare: plantswithonlyphysical-chemicaltreatment;\nTheBODrepresentstheamountofoxygenneededbybacteriatodegradetheorganicmatteranditisfunctionoftheorganicmatterconcentrationandofthedegradationrate. plantswithadditionalbiologicalreactor(forbetterorganicmatterremoval),whichcanbeoftwomainsub-type,dependingonthesortofgrowthofmicroorganisms[5]:–suspendedgrowth:withthemicroorganismsmixedwiththewastewateranddispersedintheformoffreecellsorofbioocks(activatedsludgereactors).–attachedgrowth:withthemicroorganismsanchored,intheformofbiolm,toinertsurfaces(biological-lmreactors).TheworkofthepaperfocusesonWWTPswithactivatedsludge,whichisnowthemostcommoncaseintheEuropeanUnion.2.2Rule-basedexpertsystemsRule-BasedExpertSystems(RBESs)areadvancedcomputerpro-gramswhichemulate,ortryto,thehumanreasoningandproblem-solvingcapabilities,usingthesameknowledgesources,withinapar-ticulardiscipline[23][26][9].RBESsalwayspossesscertainheuris-ticsthatformthestaticknowledge-base,andsomeinferenceandsearchprocesses.TheproblemsaddressedwithRBESsareverycom-plexandrelatedtospecicdomains,andtheywouldusuallyneedaveryexperthuman(i.e.,agreatamountofknowledge)tobesolved7.Afewexamplesofreal-worldgeneralapplicationsofRBESsarethefollowingones: decisionsupportfornaturalresourcesmanagement[18], datamanagementinforestry[31]. petrochemical-plantcontrol[1], dynamic-processmonitoringanddiagnosis[20], WWTPtime-seriesanalysis[33], controlofsun-poweredsystems[38].ThemaincomponentsofRBESsare:staticknowledge-base(orlong-timememory),database(orworkingmemoryorshort-timememory),inferenceengine,userinterface,auto-explanationmod-ule,strategymodule,knowledge-engineerinterfaceandon-linesen-sor/effectorsinterface.Typically,theknowledgecontainedinthehistoricaldataisen-codedinthestaticknowledge-baseintheformofrulesoraxioms,viaaknowledge-acquisitionprocess.Therulesallowthesystemtodeducenewresultsfromaninitialsetofdata(premises).Aruleisbasicallyrepresentedbythefollowingcode:IFconditionsTHENactionsThereasoningmethod(inferenceengine)mayuseforwardchain-ing,backwardchainingoracombinationofbothofthem.Forward-chainingreasoning(deduction)startsfromtheinputdatatowardsthenalconclusions,deducingnewfactsfrompreviousones.Backward-chainingreasoning(induction)isguidedbytheconclusionstowardstheinputdata(commonlyprovidedbytheuser).Thankstotheircharacteristics,RBESshavebeenwidelyandsuc-cessfullyappliedtoenvironmentmanagement,supervisionandcon-trol[11][36][30][14][41].2.3Experientialknowledgeandcase-basedreasoning(CBR)CBRisbothaparadigmforcomputer-basedproblemsolversandamodelofhumancognition.Thecentralideaisthattheproblemsolverreusesthesolutionfromsomepastcasetosolveacurrentproblem.ItmayevenhappenthattheRBESalgorithmic-powercoulddosomespecialtasksthatthehumanone(themind)cannotdointhegreatmajorityofthecases.8-3 2.3.1CBRasacomputerprogramparadigmAsaparadigmforcomputer-basedproblemsolvers,oneofthead-vantagesofCBRsystemsisthattheyimprovetheirperformance,becomingmoreefcient,byrecallingoldsolutionsgiventosimi-larproblemsandadaptingthemtotthenewproblems.Inthiswaytheydonothavetosolvenewproblemsfromscratch8.Thememo-rizationofpastproblems/episodesisintegratedwiththeproblem-solvingprocess,whichthusrequirestheaccesstopastexperiencetoimprovethesystem'sperformance.Additionally,case-basedrea-sonersbecomemorecompetentduringtheirfunctioningovertime,sothattheycanderivebettersolutionswhenfacedwithequallyorlessfamiliarsituationsbecausetheydonotrepeatthesamemistakes(learningprocess).ThebasicstepsinCBRare:1.Introducinganewproblem(orsituation)intothesystem.2.Retrievingapastcase(aproblemandsolution)thatresemblesthecurrentproblem.Pastcasesresideincasememory.Casememoryisadatabasethatcontainsrichdescriptionsofpriorcasesstoredasunits.Retrievingapastcaseinvolvesdeterminingwhatfeaturesofaproblemshouldbeconsideredwhenlookingforsimilarcasesandhowtomeasuredegreesofsimilarity.ThesearereferredtoastheIndexingProblemandtheSimilarityAssessmentProblem.3.Adaptingthepastsolutiontothecurrentsituation.Althoughthepastcaseissimilartothecurrentone,itmaynotbeidentical.Ifnot,thepastsolutionmayhavetobeadjustedslightlytoaccountfordifferencesbetweenthetwoproblems.ThisstepiscalledCaseAdaptation.4.Applyingtheadaptedsolutionandevaluatingtheresults.5.Updatingcasememory.Iftheadaptedsolutionworks,anewcase(composedoftheproblemjustsolvedandthesolutionused)canbeformed(directlearning).Ifthesolutionatrstfails,butcanberepairedsothefailureisavoided,thenewcaseiscomposedoftheproblemjustsolvedandtherepairedsolution.Thisnewcaseisstoredincasememorysothatthenewsolutionwillbeavailableforretrievalduringfutureproblemsolving.Inthisway,thesystembecomesmorecompetentasitgainsexperience.Updatingcasememoryincludesdeletingcases(forgetting),too.ThisstepisalsopartoftheIndexingProblem.Notallcase-basedproblemsolversuseallofthesteps.Insome,thereisnoadaptationstep;theretrievedsolutionisalreadyknowntobegoodenoughwithoutadaptation.Inothers,thereisnomem-oryupdatestep;thecasememoryismatureandprovidesadequatecoverageforproblemsinthedomain.2.3.2CBRandwastewaterenvironmentIntheWWTPdomain,CBRhasbeenusedfordesigningmoresuit-ableoperationstotreatasetofinputcontaminants[29]andforsuper-vision[36][37].Inthiscontext,thecasesstoredinthecaselibraryarerealWWTPoperatingstates,whicharelearnedinsuchawaythatitispossibletoreemploythemtosolvefuturetasks.Acasein-corporatesthefollowingsetoffeatures:anidentier,thesituationdescription,thesituationdiagnosis,theactionplan,thederivation(fromwherethecasehasbeentaken/adapted),thesolutionresult(success/failure),autilitymeasure,adistance/similarityvalue.2.3.3CBR'sproblemsIngeneral,case-basedreasoningprovedtobeagoodchoiceforexperiential-knowledge(specic-knowledge)management.ButNon-blindproblem-solvingstrategy.CBRhasthebasicproblemthatitcannotworkaloneifthereisnoavailableexperience,suchasinthecaseoftheinitialrunningpe-riodofatreatmentplant.Ithastobecombined,forinstance,witharule-basedoranontology-basedsystem(general-knowledgeman-agers)sothatitcanworkasareasoningcomponentintheoverallcontrolandsupervisionofWWTPs.AnintegrationofdifferentAImethodsisneeded,thatincludesthemanagementofqualitativein-formation(e.g.microbiologicaldescriptors,inthecaseofwastewatertreatment),experts'intelligenceandexperientialknowledge.2.4Ontologies2.4.1AIdenitionsAIliteratureisfullofdifferentdenitionsofthetermontology.Eachcommunityseemstoadoptitsowninterpretationaccordingtotheuseandpurposesthattheontologiesareintendedtoservewithinthatcommunity.\rOneoftheearlydenitions:'Anontologydenesthebasictermsandrelationscomprisingthevocabularyofatopicareaaswellastherulesforcombiningtermsandrelationstodeneextensionstothevocabulary.'[32]\rAwidelyuseddenition(Gruber):'Anontologyisanexplicitspecicationofaconceptualization.'[24]\rAnelaborationofGruber'sdenition:'Ontologiesaredenedasaformalspecicationofasharedconceptualization.'[6]2.4.2Ontologicalknowledge-basesKnowledgebases(KBs)runthroughaspectrumfromsimplecol-lectionsoffrequently-askedquestions(FAQs)tocomplexsystemspoweredbyAIengines.Historically,thetermknowledgebasereferstoabaseofexpertinformationandanswerstocommonquestions.Byprocessingitsknowledgebaseusingrulescalledheuristics,anexpertsystemcanrespondtoaseriesofquestionsandchoices,andsolveaproblemasthoughtheuserweredealingwithahumanexpertinaparticulareld.Today,thetermknowledgebasehasdevelopedatleasttwosecondmeanings:\rOneinthecontextoftheworldwideweb.Inthisdomain,aknowl-edgebaseissimplyabaseoftechnicalinformationoranswerstocommonproblems,oftenrelatedtoaparticularsystemorproduct.Thesewebknowledge-bases(WKBs)maybeprovidedasacus-tomerserviceonacorporatewebsite,ortheymaybedevelopedbyknowledgeengineersforknowledgeworkerswithinaninstitu-tionorcompany.WhilesomeoftheseknowledgebasesareusedbyexpertsystemsorotherAIsystemstosolveproblems,mostarejustpartofsimplersearchengines,likethosegenerallyusedtosearchtheWeb.\rOneinthecontextofAI.InthispaperwerefertoKBsonlyas:–ontologicalknowledge-bases(OKBs):inthedomainofAI-ontologies,aKBisthecomputer-readabletranslationofanon-tology;theseOKBsaresometimespartofamoregeneralexpertsystem;–knowledgebasesofrule-basedexpertsystemsorstaticknowledge-bases(SKBs);–knowledgebasesofcase-basedreasonersordynamicknowledge-bases(DKBs).InAI,KBswereborntohelpinknowledgereuseandsharing:'reuse'meansbuildingnewapplicationsassemblingcomponents8-4 alreadybuilt,while'sharing'occurswhendifferentapplicationsusethesameresources.Reuseandsharingpresentthefollowingadvantage:needforlessmoney,lesstimeandlessresources.2.4.3KnowledgesharingandreuseWhensharingknowledge,itispossibletocomeacrossproblemsrel-ativeto:theconceptualizationmethod[22],thesharedvocabulary(e.g.,librariesofontologies),theformattoexchangeknowledge(e.g.,KIF(KnowledgeInter-changeFormat)),andthespeciccommunicationprotocol(e.g.,KQML(KnowledgeQueryManipulationLanguage)externalinterface).Whenreusingknowledge,themostcommonproblemsconcern:theheterogeneityofknowledge-representationformalismsandoftheimplementationlanguages(workedoutbytranslators),thelexicon,thesemantics,synonymsandhiddenassumptions(workedoutbytheveryon-tologies),andthelossofcommon-senseknowledge(addressedbyanintegrationofvariousAIparadigms,suchasontologies,naturallanguagepro-cessingandmachinelearning,withcognitivescience)[21].2.4.4Ontologydevelopment:fromarttounderstoodengineeringprocessEvenifitisnowwidelyrecognizedthatconstructingontologies,ordomainmodels,isanimportantstepinthedevelopmentofKBSs,whatislackingisaclearunderstandingofhowtobuildontologies.However,thereexistsasmallbutgrowingnumberofmethodologiesthatspecicallyaddresstheissueofthedevelopmentandmainte-nanceofontologies.Inthissectionwepresent,amongtheprojectswhichgointhedirectionofprovidingthesemethodologies,theonewhichismostrelatedtotheontologyoftheDAI-DEPUR+system:theOntolinguaProject.Foracomprehensivesurveyoftheworkwhichhasbeendonesofar,see[27].OntolinguaTheguidesfortheuseoftheOntolinguaOntologyServer[15][16][19]containadviceondeveloping,browsing,main-tainingandsharingontologiesthroughtheServer.TheOntolingualanguageisbasedonthesyntaxandsemanticsofKIF.OneofthemainbenetsinusingtheOntolinguaserveristheaccessitprovidestoalibraryofpreviouslydenedontologies.Thislibraryextendsasdevelopersaddnewontologiestotherepository.OntologyconstructioninOntolinguaisbasedontheprincipleofmodulardevelopment.Ontologiesfromthelibrarycanbere-usedinfourdifferentways:1.inclusion:ontologyAisexplicitlyincludedinontologyB.Thevo-cabularyofontologyAistranslatedintothevocabularyofontol-ogyB.ThistranslationisappliedtotheaxiomsofontologyA,too,andthetranslatedaxiomsareaddedtoontologyB[16].Multipleinclusionissupported.2.polymorphicrenement:adenitionfromanontologyisincludedinanotherontologyandrened.Forexample,theBiological-Living-Objectclass,denedinUpperCycontology,canbein-cludedinWaWOontologyandextendedtoadmitBacteria,andincludedinontologyBandextendedtoadmitaliens.3.restriction:arestricted(byaxioms)versionofoneontologyisin-cludedinanother.4.cyclicinclusion:asontologyinclusionistransitive,situationssuchasthefollowingareallowed,evenifnotrecommended:ontologyAisincludedinontologyB,ontologyBisincludedinontologyCandontologyCisincludedinontologyA.Thesedistinctionsareveryusefulinthere-useofontologies,butthespecicationoftherelationshipsamongontologiesisprobablynotcomplete[27].Ontolinguaisthedefactostandardmeansofim-plementingontologiesalthoughamorecomprehensivemethodologyneedstobeusedinconjunctionwiththeServer.OneofthemaineffortsoftheOntolinguaprojectconcernstherepresentationofuncertainknowledgewithinanontology.TheOn-tolinguarepresentationlanguageresultingfromthisworkenablesontologiestocontainrichlytextureddescriptionsthatincludeuncer-tainty,arestructuredintomultipleviewsandabstractions,andareexpressedinagenericrepresentationformalismoptimizedforreuse.TheOntolingualanguageusestheKnowledgeInterchangeFormat(KIF)asacore.Itisacomputer-interpretabledescriptionlanguageandenableseasyon-linecollaborativeconstructionofontologies[24].(http://ontolingua.stanford.edu/)Withrespecttoontologyeditors,thereareanumberofmoreorlessgenericeditorstocreateandmanageontologies.TheStanfordOntolinguaOntologyEditor(StanfordKSLNetworkServices9)isthemoststandardeditortocreateontologies.2.4.5OntologiesandtheenvironmentEnvironmentalontologiesarejustinstantiationsofthegeneralon-tologyconceptwhichassistinunderstandingadomainrelatedtothenaturalenvironmentandinmodelingtheprocessesinvolved.NoontologyapplicationexistsyetintheeldofWWTPsandnoon-tologymodelingtheevolutionofmicrobiologicalsystemshasbeendened.WethinkthattherepresentationalpowerofontologiescanbeexploitedtodeepentheknowledgeaboutthemicroorganismsofWWTPactivated-sludgeandthewastewaterdomainingeneral,andcanbeintegratedtogetherwithotherreasoningmethodstobetterthewholesupervisionofWWTPs.2.5Environmentaldecisionsupportsystems(EDSSs)Whendealingwithproblemswhichhaveanegativeimpactontheenvironment,therearequestionsthatmanagersinthepublicorpri-vatedomainhavenotthetimeortheinclinationtoconsiderand,fur-thermore,theymaynothavesufcientknowledgeofenvironmentalissuestocarryoutanassessmentinanythingotherthananentirely'adhoc'manner.Thus,EDSSsarecalledfor.AnEDSSisanintegratedKBS,appliedtoanenvironmentalissue,thatreducesthetimeinwhichdecisionsaremadeandimprovestheconsistencyandqualityofthosedecisions[25].Inthissection,wediscusswhichfeaturesanEDSSshouldinclude.AnEDSSshouldincludethefollowingfeatures:Theabilitytoassisttheuserduringproblemformulation,thatis,decidingwhichobjectivesneedtobereached,andwhenandhowthedifferentavailabletoolshavetobeapplied.Astructuredframework,whichdrawsinformationfromtheuserandtheenvironmentalsystemaboutdomain-characteristicsandhttp://www-ksl-svc.stanford.edu:5915/&service=frame-editor8-5 processesinalogicalmanner.Thisframework,besidesacquiringthedomainknowledge,hastobeabletoorganizeandrepresentit.Specicknowledge-basespertinenttothetypeofdomainbeingconsideredortotheprocessbeingcarriedoutatthesite.Theseknowledgebasescontaindataonenvironmentalparametersandprocessesthatarerelevanttothedomain(e.g.whatprocessesarerequiredtomanufactureaparticularproduct;whattoxicmaterialsareusedintheprocesses;whichkindsofphysical,chemicalandbiologicalsamplesneedtobecollected;whichistherelativeim-portanceofthefeaturesinplay;whicharetherequirementsofthelocallegislation).Ageneralenvironmentalknowledgewhichisusedtodeducetherelativesignicanceofdifferentenvironmentalimpactsgivenap-propriatedataaboutthespecicdomainandprocesses.Amoduletopresenttheanalysis'resultsinauser-friendlyman-ner.Theabilitytoassisttheuserduringtheinterpretationoftheresultsandtheselectionofthesolution.Thiscanbedonebyidentifyingthesignicantfeaturesoftheanalysis'resultsandevaluatingtheirimpactwithrespecttothetaskbeingperformed.3THEDAI-DEPUR+ENVIRONMENTALDECISION-SUPPORTSYSTEMInthissectionwedescribetheDAI-DEPUR+environmentaldecision-supportsystem(EDSS)forwastewatertreatmentplants.TheDAI-DEPUR+system,asexplainedindetailbelow,includesanontologywhichhelpstomodelthewastewatertreatmentpro-cess,payingaspecialattentiontothemanagementofthequalitativeknowledge,thatistheenvironmentalinformationonmicroorganismpresence.Aswellashelpingtomodelthedomain,theontologyaddsnewcapabilitiestotheEDSS,suchassupportofcausalreasoning,prediction,andsemi-automaticgenerationofastaticKB.TheDAI-DEPUR+systemhasanarchitectureinwhichseveralarticialintelligencetechniquesintegrateandoperateinrealtime.Ofparticularinterestistheintegrationoftheontologyfortherep-resentationofthewastewatertreatmentprocess.TheDAI-DEPUR+systemisbuilttomanagespecicWWTPs,buttheontologicalrep-resentationofthedomainwillmakeeasieritsportabilitytowardsotherWWTPsandotherdomains.TheDAI-DEPUR+systemde-rivesfromtheDAI-DEPURsystem[36].ItisitsdirectevolutionandisconstantlyunderdevelopmentinrelationwiththeresearchoftheKnowledgeEngineeringandMachineLearning(KEML)groupatUPC.Indeed,theDAI-DEPUR+systemaimstogoastepfurtherincompletingthecomprehensionofWWTP-microorganismsthroughtheuseoftheontologyandexploitingthedataonactivatedsludge.InthissectionweexplainthearchitectureofDAI-DEPUR+andits3layers:perception,diagnosisanddecisionsupport.3.1ArchitectureThearchitectureofthesystemhasamodulardesign,toimprovemodiability,understandabilityandreliability.Itbasicallyfollowsastandardverticaldecompositionapproach10:adivisionismadeintomanyspecializedsubsystems,suchasperception,diagnosis,model-ing,planning,executionandeffector-controlmodules.Fig.1containsablockdiagramofthetop-leveldecompositionofthisarchitecture.Thesystemreceivesrawdatafromthesensorsandthelaboratory,andemitscommandstothesensorsandeffectors.TheForthedenitionandapplicationofhorizontalandverticaldecomposition,see[8]and[28].Figure1.Top-leveldecompositionofDAI-DEPUR+.actioncomponenttakestheoutputoftheperceptioncomponentasinputanditistheonewhichgeneratescommandstoboththesen-sorsandeffectors.Thisdiffersfrommanyothersystems,inwhichthecontrolofthesensorsistheresponsibilityoftheperceptioncom-ponent.Exceptingcasesoffailure,thereisacontinuoussensorydatastreamfromallsensors,whichgoesdirectlyintotheperceptioncom-ponent,alongwiththeresultsoflaboratoryanalysesandthecom-mandsthatwerelastsenttotheeffectors.ThedetailedarchitectureofDAI-DEPUR+isschematizedinFig.2anditsactionmodelisthefollowingone:perception:datagatheringandknowledgeacquisition,diagnosis:reasoning,decisionsupport:prediction,evaluationofalternativescenarios,advising,actuationandsupervision.3.2PerceptionlayerTheDAI-DEPUR+systemoperatesinadomainwhichphysicallyconsistsofawastewatertreatmentplant.Inparticular,allthephysi-cal,chemicalandbiologicalmeasurementsaregatheredintreatmentplantslocatedinCatalunya.Someparametersaremeasuredon-linebysensors,whileotheronesaremeasuredoff-lineinlaboratories.3.2.1AwarenessThetimescalesofthetreatmentprocessesarelong,sothattheper-ceptionandthesupervisiondecisionseasilytbetweensamplingpoints.Manydecisionsupportsystemssimply”closetheireyes”whileatime-consumingsubsystem,suchasaplannerorareasoner,isin-vokedandthepenaltyforsuchunawarenessisthatperceptualinputsareeitherlostorstackedupforlaterprocessing.ThisisnotthecaseinDAI-DEPUR+becausetheWWTPenvironmentisveryslowlyevolvingcomparedtothespeedofthereasoningofthedecisionsup-portsystem:evenifaWWTPisatrulydynamicdomain,itneverchangestosuchextentthattheresultsofrelativelylongcalculationwouldnolongerbeuseful.Ifsomethinghappensthatrequires”im-mediate”actiononthepartofthesystem,DAI-DEPUR+isalwaysawareofit.8-6 Data baseWastewater treatment plantSensorsPhysical, chemical, biological analysesOn-linedataOff-linedataOntologyCase-basedreasoningRule-basedreasoningNumerical-control modelUSER INTERFACEUSEROn- and off-line actuators Data interpretation Diagnosis Decision supportDecision / ActuationReasoning integrationSupervisionBackground knowledgeImpasseImpasseresolutionFigure2.AviewofDAI-DEPUR+architecture.3.2.2TemporalintegrationInaWWTP,sampleintervalsrangefromafewsecondstoafewdays.Ourapproachtothetemporalintegrationofanumberofpro-cessesthatworkatdifferentratesistodeneaconstantminimum-cycletimefortheentiresystem.Thistimeisequalto1hourandateachtickofthistimetheinputsarereadorcalculated,somecom-putationisdoneandtheoutputsareset(bytheactioncomponentofDAI-DEPUR+).Ifaprocess,suchasalaboratoryanalysis,cannotcomplete(orevencannotbestarted)bythetickofthetime,eitherbecauseitsschedulingisnon-constantorbecauseitssamplinginter-valislongerthan1hourorbecausethereisfailure,itsoutputsareinferred,ifpossible,inanalternativeway(oftenjustreproducingtheoutputsoftheprevioushour)anditsexecutionisre-plannedforthefollowingtick.Onceobtained,thedataarearrangedaccordingtodifferentcrite-ria:separationsaremadebetweenphysical-chemicalandmicrobio-logicalfeatures,andbetweenquantitativeandqualitativeones[10].3.2.3PhysicalandchemicalfeaturesAmongtheavailablephysicalandchemicalfeatures,themostrel-evantonesusedbytheDAI-DEPUR+systemareselectedonthebasisofhumanexperience,traditionandutilitymeasures.Thesefea-turesarenotproblematicandtheirmodelingandapplicationbothinchemicalengineeringandarticial-intelligencesystemsarewelldocumentedinbibliography.3.2.4MicrobiologicalfeaturesThemodelingofmicrobiologicalfeaturesexistsinthescopeofbi-ologicaldisciplines,butithasnotyetbeenintegratedintoadeci-sionsupportsystemdedicatedtoenvironmentalissues,suchastheDAI-DEPUR+system.InthissectionwedescribethemethodologyfollowedintheknowledgeacquisitionrelatedtoDAI-DEPUR+[10].InaWWTP,theidenticationofthemicroorganismsexistingintheactivatedsludgeisgenerallycarriedoutinthelaboratoriesoftheplantandgeneratesqualitativeoff-linedata(e.g.,presenceofParameciaspeciesordiversityofCiliate).Usinganautomaticquan-titativeanalysisofdigitalimages,formicroorganismrecognitionandcountingisapossibilityforthefuture.Aftertheidentication,acomparativestudyofmicroorganismcommunitiesofdifferenttreatment-plantsisaccomplished,toun-derstandwhatcanbetheinuenceofbiologicalvariabilityatage-ographicallevel.Asetofmicrobiologicalfeaturesisthenselectedtobeusedbythesystem.Forahighperformancetobemaintainedthroughoutthedomain(thedifferentWWTPs),thisfeaturesetneedstobewidespreadenoughtohavearepresentationaldata-basewitharelativelyabundantnumberofinstances.Referringtoportability,theparametersavailableonlyintheminorityofthetreatmentplantsarenotveryusefulinthedevelopmentofthemainknowledge-basesofthesystem,buttheycanbeusedasspecic-domainknowledgebyspeciallydevelopedmodules.Missingandincompleteinforma-tiondoesnotrepresentaprobleminprinciple,butonlyafactorofincreasinguncertainty.3.3DiagnosislayerOncealldatahavebeeninterpreted,theuseofdiagnosticknowledge-basesbegins.Diagnosisisbasicinthedecisionmakingofwastewatertreatment.Andthediagnosislayeristheonewithmostresources8-7 allocated.TheknowledgebasesmodeltheparticularkindofWWTPfromwhichthedataarecoming.Thediagnosisisbasedondifferentreasoningmodelsandtheontology.3.3.1Knowledge-basedparadigmIntheDAI-DEPUR+systemthereareanumericalcontrolmoduleandtwoAIknowledge-baseswhich:detectwhentheplantisinanormalstateorinastandardabnor-malstate,suchasbulking,stormorfoamingstates,andcontributetomanagethegeneralwastewater-treatmentoperationinthesecases.Thisroutinemanagementiscarriedoutthroughautomatic-controlalgorithms,case-basedreasoningandrule-basedreasoning.Case-basedreasoningisoftenabletomodelalsospecicfeaturesandpar-ticularstatesofthetreatmentplant(nonstandardabnormalstates),andtolearnfrompastsituationsoccurringinthetreatmentplantit-self.Thiswouldaccountforthepotentialdifferenceinindividualtreatment-plantsduetodeviationsinparameterssuchasinow,me-teorology,neighboringindustriesandlocallife-style.3.3.2OntologyAnontologyisintegratedwiththeKBSsmentionedinsectionabove.Withthisontologyitispossibletocapture,understandanddescribetheknowledgeaboutthewholephysical,chemicalandmicrobiolog-icalenvironmentofaWWTP.Thegoaloftheintegrationoftheontologyistocreateamodelthat:1.providesasharedterminologyforthewastewaterdomainthateachagentcanjointlyunderstandanduse;e.g.,thesharedtermdescriptoruniesthetermsvariable,feature,attributeandpa-rameter,whichareusedbydifferentagentstorefertothesameconcept;2.denesthemeaningofeachterm(partofthesemantics)inapre-ciseandasunambiguousmanneraspossible;e.g.,'thetermde-scriptorreferstoattributeswhichdescribeenvironmentalcondi-tions,suchastheappearanceofmicroorganismocksorofthewatersurfaceoftheclarier';3.encodesinanenvironmentaldecision-supportsystem,forthersttime,adeepmicrobiologicalknowledge;e.g.,thetaxonomyofthemicroorganismswhichliveinWWTPs;4.linksconceptswithtaxonomic/hierarchicalrelations;e.g.,'No-cardiaisaActinomycetes';5.implementsthesemanticsinasetofaxiomsthatwillenabletheontologytoautomaticallydeducetheanswertomanyquestionsaboutthewastewaterdomain,suchascause-effectquestions;e.g.,theaxiom'ActinomycetesiscauseofFoamingsludge';6.hasaxiomswhichpermitdiagnosis-impassesolving;7.usestheOntolinguaenvironmentfordepictingconceptsinagraphicalcontext;8.willintegratewithsometemporalreasoning,basedontransitionnetworks,toobtainaqualitativesimulationoftheevolutionofWWTPstates.AxiomsWeapproachgoals5and6bydeningasetofaxioms(orrules)thatdescribewastewaterprocesses.Axiomdeductionsshouldbedeterminedbyasetofquestionsusedtodecidethecompetenceoftheontology'srepresentation.Sincetheredoesnotexistastandardfordeterminingthecompetenceofamodel,wewilldeneasetofquestionsaboutwastewaterprocessesandtheaxiomsusedtoanswerthem.BasicentitiesThebasicentitiesintheontologyarerepresentedasobjectswithspecicpropertiesandrelations.Objectsarestructuredintoataxonomyandthedenitionsofobjects,attributesandrela-tionsarespeciedaccordingtotheOntolinguaversionoftheframeontology.ThehierarchicalstructureandtheaxiomsoftheontologycanhelptodiagnosethesituationincaseofimpasseoftheotherKBSs.Theontologyisnormallystatic.Itactivatesitsinferencemecha-nisms(axioms)onlyunderspecicpetitionsfromthediagnosisinte-grator(seenextsection).Theresultoftheinferenceoftheontologyis:anansweraboutthediagnosisimpasse(e.g.:'Wehaveafoamingsituation'or'Idonothaveinformationtosolvetheimpasse'),anexplanationoftheanswer(e.g.:'Ireceivedinformationrelatedtotheanswerfromtheactivationofthefollowingaxioms...'or'Theanswerwasobtainedsearchingthefollowingclasses...').Theactivationoftheontologyalwaysmeansthattherewasanim-passeinKBSdiagnosis.Iftheansweroftheontologytothepetitionis'Idonothaveinformationtosolvetheimpasse',thenaprimaryalarmisactivated.3.3.3DiagnosisintegrationTherule-basedexpertsystem(RBES)andthecase-basedreasoningsystem(CBRS)workinparallelandtheybothproduceasoutputadiagnosisonthestateoftheplant.Thisoutputispassedtothediag-nosisintegrator,asubsystembetweenthediagnosisandthedecisionsupportlayers.GeneralintegrationschemaIfthediagnosisofthetwoKBsys-temsisthesame,itispassedtothedecisionsupportlayer.Ifthediagnosesexistandaredifferent,thesystemprioritizesasfollow:Ifthecaselibrarycontainsapredenedminimumhistoricalseriesandthecasesimilarityishigherthanapredenedvalue,thecase-basedreasoner'sdiagnosisprevails.Otherwise,therule-basedexpertsystem'sdiagnosisprevails.Incaseofimpasse(nodiagnosis),DAI-DEPUR+turnsrsttotheontologyandthen,ifitfails,totheplantmanager,demandinganoff-linediagnosisbasedontheirmicrobiologicaldeepknowledge.Thisexternalsolutionislearned.Detailedintegrationschema1.CBRSdiagnosisandRBESdiagnosis:impasse,thediagnosisintegratorturnstotheontology.2.CBRSdiagnosisandRBESdiagnosis:RBESdiagnosis-certaintya:RBESdiagnosispassedtode-cisionsupportlayer.RBESdiagnosis-certaintya:impasse,thediagnosisintegra-torturnstotheontology.3.CBRSdiagnosisandRBESdiagnosis:CBRScase-similarityb:CBRSdiagnosispassedtodecisionsupportlayer.8-8 CBRScase-similarityb:impasse,thediagnosisintegratorturnstotheontology.4.CBRSdiagnosisandRBESdiagnosis:CBRScase-similarityb:CBRSdiagnosispassedtodecisionsupportlayer.CBRScase-similarityb:–RBESdiagnosis-certaintya:RBESdiagnosispassedtode-cisionsupportlayer.–RBESdiagnosis-certaintya:impasse,thediagnosisinte-gratorturnstotheontology.Example(case1)WehaveacertainperceptionstateA.Thecase-basedreasoningsystem(CBRS)andtherule-basedexpertsystem(RBES)activate.TheCBRSndsacasesimilartostateA.Thesimilarityvalueis0.1anditislessthantheminimumacceptablevalueb(e.g.b=0.2).ThereforethereisnodiagnosisoutputfromtheCBRS.TheRBESndsnorulesleadingtoadiagnosisstartingfromstateA.ThereforethereisnodiagnosisoutputfromtheRBES.ThediagnosisintegratoracknowledgesacaseofmissingdiagnosisfromtheKBSsandsendapetitiontotheontologywiththedescrip-tionofstateA.Theoutputoftheontologyis:Answer='Wehaveafoamingsitu-ation',Explanation='Ireceivedinformationrelatedtotheanswerfromtheactivationofthefollowingrelation(NocardiaisaActi-nomycetes)andthefollowingaxioms(ActinomycetesiscauseofFoamingsludge)'.Comment:StateAischaracterizedbyastrongpresenceofNocar-diabacterium,butthisbacteriumhasnodirectrelation(inthestaticknowledge-baseoftheRBES)toanystateoftheWWTP.EvenintheontologytheNocardiaclasshasnolinktoanystateoftheWWTP,butitsparentclass(Actinomycetes)hasacause-effectlinktothegen-eralstateoftheWWTP'Foamingsludge'.OnereasonforthiscouldbethatNocardiaisnotcausingfoaming,butanother(notdetected)bacteriumofthesametaxonomicclassis.Thediagnosisintegratorreceivesthediagnosisfromtheontologyandpassittothedecisionsupportlayer.3.4DecisionsupportlayerEventually,wedescribethesupervisoryleveloftheDAI-DEPUR+environmentaldecision-supportsystem.Oncethediagnosesofthereasoners(case-basedandrule-based)andpossiblyoftheontologyforthemanagementofthewastewatertreatmenthavebeeninte-grated,itselectsanactuation.RealtimeThislayerrunsalwaysinrealtimeanditisveryro-bustinthissense,simplybecauseofthefactthattheentiresystem'sminimum-cycletimeisverylongwithrespecttothecalculationsdonebythedecisionsupportlayer.3.4.1PredictionTheresultofdiagnosisintegration(carriedoutamongcase-basedreasoning,rule-basedreasoningandtheontology)servesasinputforthepredictionphase.Asubsystem,basedontransitionnetworks,pre-dictsvariousalternativeevolutionsofthestateoftheWWTP.Asec-ondsubsystemevaluatesthesealternatives.Theresultispassedontotheactuationselector.ActuationselectionActionstobecarriedoutareselected.Often,actionschemasarealreadyincludedinthediagnosisresult.4CONCLUSIONSANDFUTUREWORKInthispaperwepresentedtheDAI-DEPUR+decisionsupportsys-tem,integratingarule-basedexpertsystem,acase-basedreasonerandanontologicalknowledge-base.Thissystemisabletomodeltheinformationaboutawastewatertreatmentprocess.Themainim-provementsofDAI-DEPUR+systemwithrespecttoexistentsystemare:Impassesituationsinexistentsystemsaresolvedbytheontology.Whileinexistentsystemsthereisnomodelingofwastewatermi-crobiology,inthisnewsystemthemicrobiologicalcomponentismodeledbytheontology.DAI-DEPUR+presentsanovelintegrationbetweenKBSsandon-tologiesinarealworldapplication.DAI-DEPUR+facilitatesitsownportability.DAI-DEPUR+incorporatescause-effectreasoning.DAI-DEPUR+willincorporatepredictiveskills.Itwillbepossibleasemi-automaticgenerationofastaticKB.AcknowledgmentsThisresearchissupportedbytheGeneralitatdeCatalunya(grant1997FI00630UPCPG).REFERENCES[1]Alam´an,X.,Romero,S.,Aguirre,C.,Serrahima,P.,Mu˜noz,R.,L´opez,V.,Dorronsoro,J.anddePablo,E.,1992.MIP:ARealTimeExpertSystem,intheProceedingsofthe8thConf.OnArticialIntel-ligenceApplication(CAIA-92),Monterrey,California.[2]Avouris,N.M.,1995.Co-operatingKnowledge-BasedSystemsforEnvironmentalDecision-Support,inKnowledge-BasedSystems8(1)(1995),pp.39-53.[3]Baeza,J.,Gabriel,D.andLafuente,J.,1998.AnExpertSupervisorySystemforaPilotWWTP,inECAI98-W7(BESAI98)workshopnotes,pp.25-37.[4]Barnett,M.W.andAndrews,J.F.,1990.KnowledgeBasedSystemsforOperationofWastewaterTreatmentProcesses,inInstrumenta-tion,ControlandAutomationofWaterandWastewaterTreatmentandTransportSystems,Proceedingsofthe5thIAWPRCworkshop,Yoko-hamaandKyoto,pp.211-18,PergamonPress,NewYork1990.[5]Beccari,M.,1991.Principideitrattamentibiologici,inCanziani,R.(editor)Trattamentodelleacquediriuto,pp.99-113,Istitutoperl'Ambiente,Milano,Italy1991.[6]Borst,P.,Akkermans,H.andTop,J.,1997.EngineeringOntologies,inInternationalJournalofHuman-ComputerStudies46,pp.365-406,1997.[7]Branting,K.,Hastings,J.D.andLockwood,J.A.,1997.CARMA:apestmanagementadvisorysystem,inAIApplications11(1)(1997),pp.29-48.[8]Brooks,R.A.,1986.Arobustlayeredcontrolsystemforamobilerobot,inIEEEJ.Rob.Aut.(RA)2(1)(1986),pp.14-23.[9]Buchanan,B.andSmith,R.,1988.Fundamentalsofexpertsystems,inAnnualReviewsonComputerScience3(1988),pp.23-58.[10]Comas,Q.,R-Roda,I.,Ceccaroni,L.andSanchez-Marre,M.,1999.Semi-automaticlearningwithquantitativeandqualitativefeatures,inProceedingsoftheVIIIConferenceoftheSpanishAssociationforArticialIntelligence(CAEPIA-99),November16-19,1999,Murcia,Spain.[11]Cort´es,U.andSanchez-Marre,M.(editors),1998.Bindingenviron-mentalsciencesandarticialintelligence,ECAI98-W7(BESAI98)workshopnotes.[12]Damiani,A.,1991.Caratteristichedeiliquamiurbani,inCanziani,R.(editor)Trattamentodelleacquediriuto,pp.11-42,Istitutoperl'Ambiente,Milano,Italy1991.8-9 [13]Droste,R.L.,1997.Theoryandpracticeofwaterandwastewatertreatment,Wiley1997.[14]Dym,C.L.andLevitt,R.E.,1991.Knowledge-basedSystemsinEngi-neering,McGraw-Hill1991.[15]Farquhar,A.,Fikes,R.,Pratt,W.andRice,J.,1995.CollaborativeOntologyConstructionforInformationIntegration,technicalreportKSL-95-63,August1995,KnowledgeSystemsLaboratory,Depart-mentofComputerScience,StanfordUniversity,Stanford,US.[16]Farquhar,A.,Fikes,R.andRice,J.,1996.TheOntolinguaserver:atoolforcollaborativeontologyconstruction,technicalreportKSL-96-26,KnowledgeSystemsLaboratory,StanfordUniversity,Stanford,US.[17]Fedra,K.,1993.Expertsystemsinwaterresourcessimulationandoptimization,inMarco,J.B.etal.(editors)Stochastichydrologyanditsuseinwaterresourcessystemssimulationandoptimization,pp.397-412,KluwerAcademicPublishers,TheNetherlands1993.[18]Fedra,K.,1995.Decisionsupportfornaturalresourcesmanagement:models,GISandexpertsystems,inAIApplications9(3)(1995),pp.3-19.[19]Fikes,R.,Farquhar,A.andRice,J.,1997.ToolsforAssemblingModu-larOntologiesinOntolingua,technicalreportKSL-97-03,KnowledgeSystemsLaboratory,StanfordUniversity,Stanford,US.[20]Finch,F.E.,Oyeleye,O.O.andKramer,M.A.,1990.ARobustEvent-OrientedMethodologyforDiagnosisofDynamicProcessSystems,inComputers&ChemicalEngineering14(12)(1990),pp.1379-96.[21]G´omez-P´erez,A.,1998.Knowledgesharingandreuse,intheHand-bookofAppliedExpertSystems,pp.10.1-10.36,CRCPress1998.[22]G´omez-P´erez,A.,Fern´andez,M.anddeVicente,A.,1996.Towardsamethodtoconceptualizedomainontologies,intheproceedingsoftheworkshoponOntologicalEngineering,ECAI'96,Budapest,Hungary,pp.41-52.[23]Gonz´alezandDankel,1994.Theengineeringofknowledge-basedsys-tems,Prentice-Hall1994.[24]Gruber,T.R.,1993.ATranslationApproachtoPortableOntologies,inKnowledgeAcquisition5(2),pp.199-220,1993.[25]Guariso,G.andWerthner,H.(editors),1989.EnvironmentalDecisionSupportSystems,EllisHorwood-Wiley1989.[26]Jackson,P.,1990.Introductiontoexpertsystems,AddisonWesley1990.[27]Jones,D.M.,Bench-Capon,T.J.M.andVisser,P.R.S.,1998.Method-ologiesforOntologyDevelopment,IT&KNOWS-InformationTechnologyandKnowledgeSystems,15thIFIPWorldComputerCongress,Vienna(Austria)andBudapest(Bulgaria).[28]Kaelbling,L.P.,1990.Anarchitectureforintelligentreactivesystems,inAllen,J.,Hendler,J.andTate,A.(editors)ReadingsinPlanning,pp.713-28,MorganKaufmann1990.[29]Krovvidy,S.andWee,W.G.,1991.Wastewatertreatmentsystemsfromcase-basedreasoning,inMachineLearning10(1991),pp.341-63.[30]Mason,C.(editor),1995.WorkshoponArticialIntelligenceandtheEnvironment,inIJCAI-95WorkshopProgramWorkingNotes.[31]Matwin,S.,Charlebois,D.,Goodenough,D.G.andBhogal,P.,1995.Machinelearningandplanningfordatamanagementinforestry,inIEEEExpertSystems10(5)(1995).[32]Neches,R.,Fikes,R.,Finin,T.,Gruber,T.,Patil,R.,Senator,T.andSwartout,W.R.,1991.EnablingTechnologyforKnowledgeSharing,inAIMagazine(winter1991),pp.36-56.[33]Novotny,V.,Jones,H.,Feng,X.andCapodaglio,A.G.,1990.TimeSeriesAnalysisModelsofActivatedSludgePlants,inWaterScience&Technology23(4-6)(1990),pp.1107-16.[34]Ortolano,L.,LeCoeur,G.andMacGilchrist,R.,1990.ExpertSys-temforSewerNetworkMaintenance:ValidationIssues,inJournalofComputinginCivilEngineering4(1)(1990),pp.37-54.[35]Patry,G.andChapman,D.(editors),1989.Dynamicmodellingandexpertsystemsinwastewaterengineering,LewisPublisher,Chelsea,MI,USA1989.[36]Sanchez-Marre,M.,1995.DAI-DEPUR:anintegratedsupervisorymulti-levelarchitectureforwastewatertreatmentplants,Ph.D.The-sis,SoftwareDepartment,UPC,Barcelona,Spain.[37]Sanchez-Marre,M.,Cort´es,U.,R-Roda,I.andPoch,M.,1998.CaseLearningthroughaSimilarityMeasureforcontinuousdomains,inECAI98-W7(BESAI98)workshopnotes,pp.39-53.[38]Sanz,R.,Aguilar,J.,Sierra,C.,God´o,L.andOllero,A.,1989.Adap-tivecontrolwithasupervisorlevelusingarule-basedinferencesys-temwithapproximatereasoning,inArticialIntelligenceinScienticComputation:towards2ndgenerationsystems,IMACS,1989.[39]Serra,P.,1993.Developmentofaknowledge-basedsystemforcontrolandsupervisionofurbanwastewatertreatmentplants,Ph.D.Thesis,ChemistryDepartment,UAB,Barcelona,Spain.[40]Shepherd,A.andOrtolano,L.,1996.Water-supplysystemoperations:Critiquingexpert-systemapproach,inJ.OfWaterResourcesPlanningandManagement122(5)(1996),pp.348-55.[41]Stephanopoulos,G.,1990.Articialintelligenceinprocessengineer-ing:currentstateandfuturetrends,inComputers&ChemicalEngi-neering14(1990),pp.1259-70.[42]Wright,J.,Wiggins,L.,Jain,R.andJohnKim,T.(editors),1993.ExpertsystemsinEnvironmentalPlanning,Springer-Verlag,Berlin-Heidelberg-NewYork1993.[43]Yang,C.T.andKao,J.J.,1996.Anexpert-systemforselectingandse-quencingwastewatertreatmentprocesses,inWaterScience&Tech-nology34(3-4)(1996),pp.347-53.[44]Zhu,X.X.andSimpson,A.R.,1996.ExpertSystemforWaterTreat-mentPlantOperation,inJournalofEnvironmentalEngineering122(9)(1996),pp.822-29.8-10