Integration of rulebased expert system casebased easoner and an ontological kno wledgebase in the wastewater domain Luigi Ceccar oni Abstract
175K - views

Integration of rulebased expert system casebased easoner and an ontological kno wledgebase in the wastewater domain Luigi Ceccar oni Abstract

present an en vironmental decisionsupport system inte grating rulebased xpert system casebased reasoner and an ontological kno wledgebase This system is able to model the information about waste water treatment process through the def inition of the

Download Pdf

Integration of rulebased expert system casebased easoner and an ontological kno wledgebase in the wastewater domain Luigi Ceccar oni Abstract




Download Pdf - The PPT/PDF document "Integration of rulebased expert system c..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.



Presentation on theme: "Integration of rulebased expert system casebased easoner and an ontological kno wledgebase in the wastewater domain Luigi Ceccar oni Abstract"— Presentation transcript:


Page 1
Integration of rule-based expert system, case-based easoner and an ontological kno wledge-base in the wastewater domain Luigi Ceccar oni Abstract. present an en vironmental decision-support system inte grating rule-based xpert system, case-based reasoner and an ontological kno wledge-base. This system is able to model the information about waste water treatment process through the def- inition of the basic terms and relations comprising the ocab ulary of the waste water treatment area. Furthermore, this management system optimizes the operation of waste water treatment by mor

eliable management and making easier its self- portability INTR ODUCTION The general issues we would like to address in the paper are: optimizing waste water treatment operation by mor eliable management and making easier the portability of the management system. 1.1 Ph ysical en vir onment astewater purification Contamination le els of waters con- stantly increase due basically to industrial de elopment and to the increase of population density in certain zones. aste waters, either industrial or urban, ha to be decontaminated until an adequate le el, so that the could be poured to the

surrounding hydric medium without causing problems of en vironmental deterioration. or that, there are waste water tr eatment plants (WWTPs) with techniques of physical-chemical and biological treatment. Ev ery treatment process carries, in greater or minor measure, eco- nomic and en vironmental costs as for that it generates another kind of waste that needs, in turn, of other elimination techniques. In this sense, for the case of waste waters, the use of biological processes of purification is fa ored er the physical-chemical ones. Biologi- cal processes, in general, almost do not

consume reagents, the are more ef ficient, the do not generate either gases or noxious sludges, and the are responsible of higher production of sludges that can be used (not always) in other producti processes (e.g. as fuel, fertilizer and filling material). aste water treatment plants with biological processes are the physical en vironment modeled by our system. The general operation of WWTP always includes arious internal pre-designed standard units, whose sub-operation is optimized to accomplish single task Uni versitat Politecnica de Catalunya, Departament de Llenguatges Sis-

temes Informatics, Campus Nord, Modul C6, Jordi Girona 3, 08034 Barcelona, Spain, email: luigic@lsi.upc.es task in this context is generally the remov al remediation of contaminant substances or pathogenic microor ganisms. [13]. Each sub-operation usually has ef fects on other do wnstream treatment processes, and tradeof fs between increasing the ef ficienc of one process or another are necessary taking into account major constraints such as water characteristics, ef fluent quality and costs of each operation. 1.2 Softwar en vir onment The system we propose (named AI-DEPUR+) recei

es on-line in- puts from sensors all er the WWTP as well as of f-line inputs from the WWTP laboratories and human operators. The system uses its in- ternal kno wledge-base and inference mechanisms to process and understand this information, to diagnose the ongoing WWTP-state, and to predict the olution of that WWTP state. Ev entually the out- put of the system is represented by statements about actions to be taken, or statements to support human decisions in future actuations, or direct control signals to WWTP de vices in order to maintain the plant working correctly In case of diagnosis

impasse, AI-DEPUR+, before turning to the plant manager will try to solv the problem xploiting the connec- tion, in the ontology between data and states of the WWTP 1.3 Motivations The process of waste water treatment is so comple that it is dif fi- cult to de elop reliable supervisory technology based only on chemical-engineering classic-control approach. No wadays the use of Artificial-Intelligence (AI) systems seems to be necessary in order to obtain better results in waste water management. Rule-based xpert systems (one of the broadly applied paradigms of AI) pro ed able to

cope with some kno wn dif ficulties and to face se eral WWTP-domain problems, en if the are not the definiti solution to the treatment problem as whole. On the other hand, lar ge, multifunctional, ailable ontologies would significantly im- pro current xpert systems and tutoring systems because the con- tain the broad kno wledge of domain required to perform multiple Input/output de vices are any of arious de vices used to enter information and instructions into the AI-DEPUR+ system for storage or processing and to deli ver the processed data to human operator or in some

cases, machine controlled by the system. Such de vices comprise sensors and ef fectors. Apparatus of this kind with direct connection to AI-DEPUR+ central processing unit is said to be on-line; peripheral equipment working independently of it is termed of f-line. Both static (rule-based), dynamic (case-based) and ontological kno wledge bases. Inference is the process of drawing conclusions about particular parameter of the domain.
Page 2
tasks and to xplain domain kno wledge from multiple vie wpoints. lot of ontologies are no wadays being uilt in man research centers around the

world, ut fe of them are specialized in biology or ecology let alone waste water management and WWTP microbiol- ogy great impro ement in addressing this kind of problems can come from the inte gr ation of dif ferent modeling and reasoning sys- tems, such as specific ontologies rule-based easoning case-based easoning and eactive planning The architecture we adopt (D AI- DEPUR+) inte grates two kno wledge-based systems and an ontology and is fle xible enough to deal with the comple xity of the waste water treatment process, gi en an adequate amount and kind of data. In the AI-DEPUR+

architecture, with the embedded ontology for the waste water treatment, the representation of deeper kno wledge of the domain is permitted and the olution of WWTP-microor ganism communities can be taken into account. In this way the management of biological problems arising in treatment plant is more ef fecti e. or the first time, by the inte gration of an ontology with two kno wledge-based systems, it will be possible to capture, understand and describe the kno wledge about the whole physical, chemical and microbiological en vironment of waste water treatment plant. The basic terms and

axioms of the ontology will entail model of waste water domain and classification of the microor ganisms according to kno wn biological taxonomy and will include com- plete description of the microor ganisms themselv es (physical aspect, ab undance and beha vior in the treatment plant). The relations of the ontology will include, besides the classic hierarchical af filiations, all the interesting bindings among microor ganisms and between them and the state of the plant (diagnosis potential). 1.4 General overview In this paper we present decision support system for the supervision

of waste water treatment plants, which is part of the kno wledge and technology needed for the rational management of water resources. start by describing (in section 2) related work on the en vironmental-domain study and the AI techniques (including on- tologies) related to the creation of en vironmental decision-support systems. In section we xplain ho the decision support system (D AI-DEPUR+) has been designed and we include description of its layered architecture. Ev entually in section the contrib utions of this work are summarized and discussed. ASTEW TER DOMAIN AND AI TECHNIQUES 2.1

astewater tr eatment pr ocess In this section we describe the general treatment process, its possible ariations, and waste water description from physical, chemical and biological point of vie astewater tr eatment pr ocess The waste water treatment pro- cess is part of the water ycle and, as such, it has direct relation with other water systems or reserv oirs. aste water treatment plants (WWTPs) recei water from the anthropic system of se wers, the someho process it, and finally the deli er this water to natural reserv oir The waste water processing is what we care about, ut we cannot

for get the two other closest components of the global water ycle just mentioned (se wers, and ri er or sea). It is on the basis of the quantity and quality of water to be treated that WWTPs are uilt, taking into account the possible fluctuations in the inflo These fluctuations can be ery important where the se werage system is not ery de eloped and therefore it is not able to damp do wn inflo peaks to wards the plant. The main objecti es in waste water -treatment research are: kno wing better the rele ant characteristics of the waste water refraining the contaminated

water from reaching the natural en vi- ronment. The fact is that continuously increasing economic and cultural pressures on freshwater resources, including pollution and xcessi use, are causing threats which are augmenting costs and multiply- ing conflicts among dif ferent users of this strate gic resource. These pressures can also impair the natural re generati functions of the ecosystems in the water ycle. wo of the main challenges in the area of general water -management are to protect the water bodies and to pro vide high quality water in suf ficient quantity at af fordable

costs. In order to achie these goals, multidisciplinary research-ef forts and actions are necessary The ery xistence of WWTPs and the research for impro ving them goes in this direction and constitutes an essen- tial element for an inte grated sustainable management of water re- sources. The objecti es of such sustainable management are to de- elop technologies to pre ent and treat pollution of water to purify water to use and re-use it rationally to enhance ef ficient treatment of waste water and to minimize en vironmental impacts from waste water treatment (including the pre ention of

potential health hazards). 2.1.1 Gener al waste water har acterization Urban waste water can be characterized in accordance with the pres- ence of dif ferent kinds of dumping, such as domestic, commer cial or industrial ones. Another important feature is the presence of pathogenic or ganisms, which can prejudice possible alternati reuse of treated water such as irrigation. There are substantially two components in waste water: human metabolic waste and discarded material. While the first compo- nent is almost changeless in nature (as it is dependent on human metabolism), the second one

depends on man parameters, such as standard of li ving, local habits and country 2.1.2 Physical components otal solids can be distinguished in suspended (sedimentable or not), colloidal and dissolv ed, and contain or ganic and inor ganic portions. The size of the solids that are present in waste water influences the sedimentation, adsorption, dif fusion, mass transfer and biochemical reactions. The temper atur of waste water depends on the typology of dumping and on the permanence time in the se wers. Except for summer months, it is higher than en vironment temperature, due to the

presence of warm water dumping from kitchens and bathrooms. The importance of waste water temperature is bound to the biolog- ical acti vity of purification in treatment plants. At more than 40 nitrification halts and temperatures higher than 50 block aero- bic digestion. emperatures lo wer than 15 inhibit the anaerobic methanogenic process, while at the nitrificant autotrophic flora stops its acti vity and at also the heterotrophic flora become in- ef fecti e. aste water color is strictly correlated to its age, its septic conditions and to the presence of

industrial dumping. The odor is as- sociated to putrescence and decomposition de gree of or ganic matter and to the presence of particular industrial waste water
Page 3
2.1.3 Chemical har acteristics Here, brief description of or ganic and inor ganic chemical descrip- tors of waste water is gi en. In general, or ganic matter is rapidly biode graded, ut part of it is not and moreo er is toxic for man WWTP microor ganisms. al- uate the content of or ganic matter the bioc hemical oxygen demand and the hemical oxygen demand (COD) are determined. The majority of toxic ef fects on

WWTP-microor ganisms gro wth are attrib utable to inor ganic matter such as heavy metals and to its interaction with other waste water materials. The nitr ogen found in waste water is of pre alent kinds: or ganic nitrogen (in getal and animal proteins), ammoniacal nitro- gen, nitrites, nitrates and elemental gaseous nitrogen. Ammoniacal nitrogen is produced during the decomposition hydrolysis of or ganic nitrogen and can come from the bacterial reduction of nitrites or directly from industrial dumping. The main kinds of phosphoru xisting in waste water are: salts of orthophosphoric acid,

polyphosphates and or ganic phosphorus. In urban waste water in general, all kinds of phosphorus are present, while, after biological treatment, there are generally only or thophosphates. Sulfur is present in the form of sulfates or sulfides. Sulfates can be reduced to sulfides by sulfate-reducer bacteria in anaerobic con- ditions. Sulfites constitute culture medium for se eral species of aerobic bacteria able to create sulfuric acid, which can cause corro- sion problems. Chlorides ha metabolic human origin (as the are contained in urine in an amount equal to 1%) or are due

to industrial-water con- trib ution. Some heavy metals in waste water are necessary in minimum amounts as microelements for WWTP microor ganisms and for aquatic life, ut the are poisonous in high concentrations. 2.1.4 Biological components basic kno wledge about the most common natural or ganisms that can be found in waste water is also necessary to control the treat- ment process. Some of these or ganisms are essential for certain pollution-remo al treatments, such as acti ated sludge. The majority of pathogenic or ganisms are part of human intestinal bacterial flora and the cannot

survi for long time in waste water In general, most of the or ganisms of human origin are banal saprophytic bacte- ria, that is or ganic-matter demolishers; the are not pathogenic and can enter biological processes without an problem [12]. 2.1.5 aste water tr eatment plants In waste water treatment plant (WWTP), the main goal is to reduce the le el of pollution of the inflo water that is to remo e, within certain limits (depending on local le gislation), abnormal amounts of pollutants in the water prior to its dischar ge to the natural en viron- ment. This can be done in number of dif

ferent ways, corresponding to dif ferent kinds of WWTP The most widespread classes of WWTP are: plants with only physical-chemical treatment; The BOD represents the amount of oxygen needed by bacteria to degrade the or ganic matter and it is function of the or ganic matter concentration and of the degradation rate. plants with additional biological reactor (for better or ganic matter remo al), which can be of two main sub-type, depending on the sort of gro wth of microor ganisms [5]: suspended gro wth: with the microor ganisms mix ed with the waste water and dispersed in the form of free cells

or of bioflocks (acti ated sludge reactors). attached gro wth: with the microor ganisms anchored, in the form of biofilm, to inert surfaces (biological-film reactors). The work of the paper focuses on WWTPs with acti ated sludge, which is no the most common case in the European Union. 2.2 Rule-based expert systems Rule-Based Expert Systems (RBESs) are adv anced computer pro- grams which emulate, or try to, the human reasoning and problem- solving capabilities, using the same kno wledge sources, within par ticular discipline [23] [26] [9 RBESs always possess certain heuris-

tics that form the static kno wledge-base, and some inference and search processes. The problems addressed with RBESs are ery com- ple and related to specific domains, and the would usually need ery xpert human (i.e., great amount of kno wledge) to be solv ed fe xamples of real-world general applications of RBESs are the follo wing ones: decision support for natural resources management [18], data management in forestry [31 ]. petrochemical-plant control [1], dynamic-process monitoring and diagnosis [20 ], WWTP time-series analysis [33], control of sun-po wered systems [38 ]. The main

components of RBESs are: static kno wledge-base (or long-time memory), data base (or working memory or short-time memory), inference engine, user interface, auto-e xplanation mod- ule, strate gy module, kno wledge-enginee interface and on-line sen- sor/ef fectors interface. ypically the kno wledge contained in the historical data is en- coded in the static kno wledge-base in the form of rules or axioms, via kno wledge-acquisition process. The rules allo the system to deduce ne results from an initial set of data (premises). rule is basically represented by the follo wing code: IF conditions

THEN actions The reasoning method (inference engine) may use forward chain- ing, backward chaining or combination of both of them. orward- chaining reasoning (deduction) starts from the input data to wards the final conclusions, deducing ne facts from pre vious ones. Backward- chaining reasoning (induction) is guided by the conclusions to wards the input data (commonly pro vided by the user). Thanks to their characteristics, RBESs ha been widely and suc- cessfully applied to en vironment management, supervision and con- trol [11 [36] [30] [14] [41]. 2.3 Experiential knowledge and

case-based reasoning (CBR) CBR is both paradigm for computer -based problem solv ers and model of human cognition. The central idea is that the problem solv er reuses the solution from some past case to solv current problem. It may ven happen that the RBES algorithmic-po wer could do some special tasks that the human one (the mind) cannot do in the great majority of the cases.
Page 4
2.3.1 CBR as computer pr ogr am par adigm As paradigm for computer -based problem solv ers, one of the ad- antages of CBR systems is that the impro their performance, becoming more ef ficient, by

recalling old solutions gi en to simi- lar problems and adapting them to fit the ne problems. In this way the do not ha to solv ne problems from scratch The memo- rization of past problems episodes is inte grated with the problem- solving process, which thus requires the access to past xperience to impro the system performance. Additionally case-based rea- soners become more competent during their functioning er time, so that the can deri better solutions when faced with equally or less familiar situations because the do not repeat the same mistakes (learning process). The basic steps in

CBR are: 1. Introducing ne problem (or situation) into the system. 2. Retrie ving past case (a problem and solution) that resembles the current problem. ast cases reside in case memory Case memory is database that contains rich descriptions of prior cases stored as units. Retrie ving past case in olv es determining what features of problem should be considered when looking for similar cases and ho to measure de grees of similarity These are referred to as the Inde xing Problem and the Similarity Assessment Problem. 3. Adapting the past solution to the current situation. Although the past case

is similar to the current one, it may not be identical. If not, the past solution may ha to be adjusted slightly to account for dif ferences between the two problems. This step is called Case Adaptation. 4. Applying the adapted solution and aluating the results. 5. Updating case memory If the adapted solution works, ne case (composed of the problem just solv ed and the solution used) can be formed (direct learning). If the solution at first fails, ut can be repaired so the failure is oided, the ne case is composed of the problem just solv ed and the repaired solution. This ne case is

stored in case memory so that the ne solution will be ailable for retrie al during future problem solving. In this way the system becomes more competent as it gains xperience. Updating case memory includes deleting cases (for getting), too. This step is also part of the Inde xing Problem. Not all case-based problem solv ers use all of the steps. In some, there is no adaptation step; the retrie ed solution is already kno wn to be good enough without adaptation. In others, there is no mem- ory update step; the case memory is mature and pro vides adequate co erage for problems in the domain.

2.3.2 CBR and waste water en vir onment In the WWTP domain, CBR has been used for designing more suit- able operations to treat set of input contaminants [29] and for super vision [36] [37]. In this conte xt, the cases stored in the case library are real WWTP operating states, which are learned in such way that it is possible to reemploy them to solv future tasks. case in- corporates the follo wing set of features: an identifier the situation description, the situation diagnosis, the action plan, the deri ation (from where the case has been taken adapted), the solution result (success

failure), utility measure, distance similarity alue. 2.3.3 CBR pr oblems In general, case-based reasoning pro ed to be good choice for xperiential-kno wledge (specific-kno wledge) management. But Non-blind problem-solving strategy CBR has the basic problem that it cannot work alone if there is no ailable xperience, such as in the case of the initial running pe- riod of treatment plant. It has to be combined, for instance, with rule-based or an ontology-based system (general-kno wledge man- agers) so that it can work as reasoning component in the erall control and supervision of WWTPs. An

inte gration of dif ferent AI methods is needed, that includes the management of qualitati in- formation (e.g. microbiological descriptors, in the case of waste water treatment), xperts intelligence and xperiential kno wledge. 2.4 Ontologies 2.4.1 AI definitions AI literature is full of dif ferent definitions of the term ontology Each community seems to adopt its wn interpretation according to the use and purposes that the ontologies are intended to serv within that community One of the early definitions: ’An ontology defines the basic terms and relations comprising the

ocab ulary of topic area as well as the rules for combining terms and relations to define xtensions to the ocab ulary [32] widely used definition (Gruber): ’An ontology is an xplicit specification of conceptualization. [24] An elaboration of Gruber definition: ’Ontologies are defined as formal specification of shared conceptualization. [6] 2.4.2 Ontological knowledge-bases Kno wledge bases (KBs) run through spectrum from simple col- lections of frequently-asked questions (F Qs) to comple systems po wered by AI engines. Historically the term knowledge base

refers to base of xpert information and answers to common questions. By processing its kno wledge base using rules called heuristics, an xpert system can respond to series of questions and choices, and solv problem as though the user were dealing with human xpert in particular field. oday the term knowledge base has de eloped at least two second meanings: One in the conte xt of the world wide web In this domain, kno wl- edge base is simply base of technical information or answers to common problems, often related to particular system or product. These web kno wledge-bases (WKBs) may be

pro vided as cus- tomer service on corporate web site, or the may be de eloped by kno wledge engineers for kno wledge workers within an institu- tion or compan While some of these kno wledge bases are used by xpert systems or other AI systems to solv problems, most are just part of simpler search engines, like those generally used to search the eb One in the conte xt of AI. In this paper we refer to KBs only as: ontological knowledge-bas es (OKBs): in the domain of AI- ontologies, KB is the computer -readable translation of an on- tology; these OKBs are sometimes part of more general xpert

system; kno wledge bases of rule-based xpert systems or static knowledge-bases (SKBs); kno wledge bases of case-based reasoners or dynamic knowledge-bases (DKBs). In AI, KBs were born to help in kno wledge reuse and sharing: reuse means uilding ne applications assembling components
Page 5
already uilt, while sharing occurs when dif ferent applications use the same resources. Reuse and sharing present the follo wing adv antage: need for less mone less time and less resources. 2.4.3 Knowledge sharing and euse When sharing kno wledge, it is possible to come across problems rel- ati to:

the conceptualization method [22], the shared ocab ulary (e.g., libraries of ontologies), the format to xchange kno wledge (e.g., KIF (Kno wledge Inter change ormat)), and the specific communication protocol (e.g., KQML (Kno wledge Query Manipulation Language) xternal interface). When reusing kno wledge, the most common problems concern: the heterogeneity of kno wledge-representation formalisms and of the implementation languages (worked out by translators), the le xicon, the semantics, synon yms and hidden assumptions (worked out by the ery on- tologies), and the loss of common-sense

kno wledge (addressed by an inte gration of arious AI paradigms, such as ontologies, natural language pro- cessing and machine learning, with cogniti science) [21]. 2.4.4 Ontology de velopment: fr om art to understood engineering pr ocess Ev en if it is no widely recognized that constructing ontologies, or domain models, is an important step in the de elopment of KBSs, what is lacking is clear understanding of how to uild ontologies. Ho we er there xists small ut gro wing number of methodologies that specifically address the issue of the de elopment and mainte- nance of ontologies. In

this section we present, among the projects which go in the direction of pro viding these methodologies, the one which is most related to the ontology of the AI-DEPUR+ system: the Ontolingua Project. or comprehensi surv of the work which has been done so far see [27]. Ontolingua The guides for the use of the Ontolingua Ontology Serv er [15] [16] [19] contain advice on de eloping, bro wsing, main- taining and sharing ontologies through the Serv er The Ontolingua language is based on the syntax and semantics of KIF One of the main benefits in using the Ontolingua serv er is the access it

pro vides to library of pre viously defined ontologies. This library xtends as de elopers add ne ontologies to the repository Ontology construction in Ontolingua is based on the principle of modular de elopment. Ontologies from the library can be re-used in four dif ferent ways: 1. inclusion: ontology is xplicitly included in ontology The o- cab ulary of ontology is translated into the ocab ulary of ontol- ogy This translation is applied to the axioms of ontology too, and the translated axioms are added to ontology [16]. Multiple inclusion is supported. 2. polymorphic refinement:

definition from an ontology is included in another ontology and refined. or xample, the Biological- Li ving-Object class, defined in UpperCyc ontology can be in- cluded in aWO ontology and xtended to admit Bacteria, and included in ontology and xtended to admit aliens. 3. restriction: restricted (by axioms) ersion of one ontology is in- cluded in another 4. yclic inclusion: as ontology inclusion is transiti e, situations such as the follo wing are allo wed, en if not recommended: ontology is included in ontology ontology is included in ontology and ontology is included in

ontology These distinctions are ery useful in the re-use of ontologies, ut the specification of the relationships among ontologies is probably not complete [27]. Ontolingua is the de facto standard means of im- plementing ontologies although more comprehensi methodology needs to be used in conjunction with the Serv er One of the main ef forts of the Ontolingua project concerns the representation of uncertain kno wledge within an ontology The On- tolingua epr ese ntation language resulting from this work enables ontologies to contain richly te xtured descriptions that include uncer tainty

are structured into multiple vie ws and abstractions, and are xpressed in generic representation formalism optimized for reuse. The Ontolingua language uses the Kno wledge Interchange ormat (KIF) as core. It is computer -interpretable description language and enables easy on-line collaborati construction of ontologies [24]. (http://ontolingua.stanford.edu/) ith respect to ontology editors, there are number of more or less generic editors to create and manage ontologies. The Stanford Ontolingua Ontology Editor (Stanford KSL Network Services is the most standard editor to create ontologies.

2.4.5 Ontologies and the en vir onment En vironmental ontologies are just instantiations of the general on- tology concept which assist in understanding domain related to the natural en vironment and in modeling the processes in olv ed. No ontology application xists yet in the field of WWTPs and no on- tology modeling the olution of microbiological systems has been defined. think that the representational po wer of ontologies can be xploited to deepen the kno wledge about the microor ganisms of WWTP acti ated-sludge and the waste water domain in general, and can be inte grated

together with other reasoning methods to better the whole supervision of WWTPs. 2.5 En vironmental decision support systems (EDSSs) When dealing with problems which ha ne gati impact on the en vironment, there are questions that managers in the public or pri- ate domain ha not the time or the inclination to consider and, fur thermore, the may not ha suf ficient kno wledge of en vironmental issues to carry out an assessment in an ything other than an entirely ’ad hoc manner Thus, EDSSs are called for An EDSS is an inte grated KBS, applied to an en vironmental issue, that reduces the time

in which decisions are made and impro es the consistenc and quality of those decisions [25]. In this section, we discuss which features an EDSS should include. An EDSS should include the follo wing features: The ability to assist the user during problem formulation, that is, deciding which objecti es need to be reached, and when and ho the dif ferent ailable tools ha to be applied. structured frame work, which dra ws information from the user and the en vironmental system about domain-characteristics and http://www-ksl-svc.stanford.edu:5915 /&service= frame-ed itor
Page 6
processes

in logical manner This frame work, besides acquiring the domain kno wledge, has to be able to or ganize and represent it. Specific kno wledge-bases pertinent to the type of domain being considered or to the process being carried out at the site. These kno wledge bases contain data on en vironmental parameters and processes that are rele ant to the domain (e.g. what processes are required to manufacture particular product; what toxic materials are used in the processes; which kinds of physical, chemical and biological samples need to be collected; which is the relati im- portance of the

features in play; which are the requirements of the local le gislation). general en vironmental kno wledge which is used to deduce the relati significance of dif ferent en vironmental impacts gi en ap- propriate data about the specific domain and processes. module to present the analysis results in user -friendly man- ner The ability to assist the user during the interpretation of the results and the selection of the solution. This can be done by identifying the significant features of the analysis results and aluating their impact with respect to the task being performed.

THE AI-DEPUR+ ENVIR ONMENT AL DECISION-SUPPOR SYSTEM In this section we describe the AI-DEPUR+ en vironmental decision-support system (EDSS) for waste water treatment plants. The AI-DEPUR+ system, as xplained in detail belo includes an ontology which helps to model the waste water treatment pro- cess, paying special attention to the management of the qualitati kno wledge, that is the en vironmental information on microor ganism presence. As well as helping to model the domain, the ontology adds ne capabilities to the EDSS, such as support of causal reasoning, prediction, and semi-automatic

generation of static KB. The AI-DEPUR+ system has an architecture in which se eral artificial intelligence techniques inte grate and operate in real time. Of particular interest is the inte gration of the ontology for the rep- resentation of the waste water treatment process. The AI-DEPUR+ system is uilt to manage specific WWTPs, ut the ontological rep- resentation of the domain will make easier its portability to wards other WWTPs and other domains. The AI-DEPUR+ system de- ri es from the AI-DEPUR system [36 ]. It is its direct olution and is constantly under de elopment in

relation with the research of the Kno wledge Engineering and Machine Learning (KEML) group at UPC. Indeed, the AI-DEPUR+ system aims to go step further in completing the comprehension of WWTP-microor ganisms through the use of the ontology and xploiting the data on acti ated sludge. In this section we xplain the architecture of AI-DEPUR+ and its layers: perception, diagnosis and decision support. 3.1 Ar chitectur The architecture of the system has modular design, to impro modifiability understandability and reliability It basically follo ws standard ertical decomposition approach 10 di

vision is made into man specialized subsystems, such as perception, diagnosis, model- ing, planning, ecution and ef fector -control modules. Fig.1 contains block diagram of the top-le el decomposition of this architecture. The system recei es ra data from the sensors and the laboratory and emits commands to the sensors and ef fectors. The  For the definition and application of horizontal and vertical decomposition, see [8 and [28 ]. Figur 1. op-le vel decomposition of AI-DEPUR+. action component takes the output of the perception component as input and it is the one which generates

commands to both the sen- sors and ef fectors. This dif fers from man other systems, in which the control of the sensors is the responsibility of the perception com- ponent. Excepting cases of failure, there is continuous sensory data stream from all sensors, which goes directly into the perception com- ponent, along with the results of laboratory analyses and the com- mands that were last sent to the ef fectors. The detailed architecture of AI-DEPUR+ is schematized in Fig.2 and its action model is the follo wing one: perception: data gathering and kno wledge acquisition, diagnosis: reasoning,

decision support: prediction, aluation of alternati scenarios, advising, actuation and supervision. 3.2 er ception layer The AI-DEPUR+ system operates in domain which physically consists of waste water treatment plant. In particular all the physi- cal, chemical and biological measurements are gathered in treatment plants located in Catalun ya. Some parameters are measured on-line by sensors, while other ones are measured of f-line in laboratories. 3.2.1 war eness The time scales of the treatment processes are long, so that the per ception and the supervision decisions easily fit between

sampling points. Man decision support systems simply ”close their yes while time-consuming subsystem, such as planner or reasoner is in- oked and the penalty for such una wareness is that perceptual inputs are either lost or stacked up for later processing. This is not the case in AI-DEPUR+ because the WWTP en vironment is ery slo wly olving compared to the speed of the reasoning of the decision sup- port system: en if WWTP is truly dynamic domain, it ne er changes to such xtent that the results of relati ely long calculation would no longer be useful. If something happens that requires ”im-

mediate action on the part of the system, AI-DEPUR+ is always ware of it.
Page 7
a r ea t hy ca , ca , ca l li ff- ea on ng ea on ng ca on l m od R TE - d ff- e ac a ec n po ec on / ea ng ti on up ac ou d kn dg Figur 2. vie of AI-DEPUR+ architecture. 3.2.2 empor al inte gr ation In WWTP sample interv als range from fe seconds to fe days. Our approach to the temporal inte gration of number of pro- cesses that work at dif ferent rates is to define constant minimum- ycle time for the entire system. This time is equal to hour and at each tick of this time the inputs are read or

calculated, some com- putation is done and the outputs are set (by the action component of AI-DEPUR+). If process, such as laboratory analysis, cannot complete (or en cannot be started) by the tick of the time, either because its scheduling is non-constant or because its sampling inter al is longer than hour or because there is failure, its outputs are inferred, if possible, in an alternati way (often just reproducing the outputs of the pre vious hour) and its ecution is re-planned for the follo wing tick. Once obtained, the data are arranged according to dif ferent crite- ria: separations are

made between physical-chemical and microbio- logical features, and between quantitati and qualitati ones [10]. 3.2.3 Physical and hemical featur es Among the ailable physical and chemical features, the most rel- ant ones used by the AI-DEPUR+ system are selected on the basis of human xperience, tradition and utility measures. These fea- tures are not problematic and their modeling and application both in chemical engineering and artificial-intelligence systems are well documented in bibliography 3.2.4 Micr obiological featur es The modeling of microbiological features xists in the scope

of bi- ological disciplines, ut it has not yet been inte grated into deci- sion support system dedicated to en vironmental issues, such as the AI-DEPUR+ system. In this section we describe the methodology follo wed in the kno wledge acquisition related to AI-DEPUR+ [10]. In WWTP the identification of the microor ganisms xisting in the acti ated sludge is generally carried out in the laboratories of the plant and generates qualitati of f-line data (e.g., presence of aramecia species or di ersity of Ciliate). Using an automatic quan- titati analysis of digital images, for microor ganism

recognition and counting is possibility for the future. After the identification, comparati study of microor ganism communities of dif ferent treatment-plants is accomplished, to un- derstand what can be the influence of biological ariability at ge- ographical le el. set of microbiological features is then selected to be used by the system. or high performance to be maintained throughout the domain (the dif ferent WWTPs), this feature set needs to be widespread enough to ha representational data-base with relati ely ab undant number of instances. Referring to portability the

parameters ailable only in the minority of the treatment plants are not ery useful in the de elopment of the main kno wledge-bases of the system, ut the can be used as specific-domain kno wledge by specially de eloped modules. Missing and incomplete informa- tion does not represent problem in principle, ut only factor of increasing uncertainty 3.3 Diagnosis layer Once all data ha been interpreted, the use of diagnostic kno wledge- bases be gins. Diagnosis is basic in the decision making of waste water treatment. And the diagnosis layer is the one with most resources
Page 8

allocated. The kno wledge bases model the particular kind of WWTP from which the data are coming. The diagnosis is based on dif ferent reasoning models and the ontology 3.3.1 Knowledge-based par adigm In the AI-DEPUR+ system there are numerical control module and two AI kno wledge-base which: detect when the plant is in normal state or in standar abnor mal state, such as ulking, storm or foaming states, and contrib ute to manage the general waste water -treatment operation in these cases. This routine management is carried out through automatic-control algorithms, case-based reasoning and

rule-based reasoning. Case- based reasoning is often able to model also specific features and par ticular states of the treatment plant nonstandar abnormal states), and to learn from past situations occurring in the treatment plant it- self. This would account for the potential dif ference in indi vidual treatment-plants due to de viations in parameters such as inflo me- teorology neighboring industries and local life-style. 3.3.2 Ontology An ontology is inte grated with the KBSs mentioned in section abo e. ith this ontology it is possible to capture, understand and describe the

kno wledge about the whole physical, chemical and microbiolog- ical en vironment of WWTP The goal of the inte gration of the ontology is to create model that: 1. pro vides shared terminology for the waste water domain that each agent can jointly understand and use; e.g., the shared term descriptor unifies the terms variable featur attrib ute and pa- ameter which are used by dif ferent agents to refer to the same concept; 2. defines the meaning of each term (part of the semantics) in pre- cise and as unambiguous manner as possible; e.g., the term de- scriptor refers to attrib utes

which describe en vironmental condi- tions, such as the appearance of microor ganism flocks or of the water surface of the clarifier’; 3. encodes in an en vironmental decision-support system, for the first time, deep microbiological kno wledge; e.g., the taxonomy of the microor ganisms which li in WWTPs; 4. links concepts with taxonomic hierarchical relations; e.g., ’No- cardia is Actinomycetes’; 5. implements the semantics in set of axioms that will enable the ontology to automatically deduce the answer to man questions about the waste water domain, such as cause-ef fect

questions; e.g., the axiom ’Actinomycetes is cause of oaming sludge’; 6. has axioms which permit diagnosis-impasse solving; 7. uses the Ontolingua en vironment for depicting concepts in graphical conte xt; 8. will inte grate with some temporal reasoning, based on transition networks, to obtain qualitati simulation of the olution of WWTP states. Axioms approach goals and by defining set of axioms (or rules) that describe waste water processes. Axiom deductions should be determined by set of questions used to decide the competence of the ontology representation. Since there does not xist

standard for determining the competence of model, we will define set of questions about waste water processes and the axioms used to answer them. Basic entities The basic entities in the ontology are represented as objects with specific properties and relations. Objects are structured into taxonomy and the definitions of objects, attrib utes and rela- tions are specified according to the Ontolingua ersion of the frame ontology The hierarchical structure and the axioms of the ontology can help to diagnose the situation in case of impasse of the other KBSs. The ontology

is normally static. It acti ates its inference mecha- nisms (axioms) only under specific petitions from the diagnosis inte- grator (see ne xt section). The result of the inference of the ontology is: an answer about the diagnosis impasse (e.g.: ’W ha foaming situation or ’I do not ha information to solv the impasse’), an xplanation of the answer (e.g.: ’I recei ed information related to the answer from the acti ation of the follo wing axioms ... or ’The answer was obtained searching the follo wing classes ... ’). The acti ation of the ontology always means that there was an im- passe in

KBS diagnosis. If the answer of the ontology to the petition is ’I do not ha information to solv the impasse’, then primary alarm is acti ated. 3.3.3 Diagnosis inte gr ation The rule-based xpert system (RBES) and the case-based reasoning system (CBRS) work in parallel and the both produce as output diagnosis on the state of the plant. This output is passed to the diag- nosis inte grator subsystem between the diagnosis and the decision support layers. General integration schema If the diagnosis of the two KB sys- tems is the same, it is passed to the decision support layer If the diagnoses xist

and are dif ferent, the system prioritizes as follo w: If the case library contains predefined minimum historical series and the case similarity is higher than predefined alue, the case- based reasoner diagnosis pre ails. Otherwise, the rule-based xpert system diagnosis pre ails. In case of impasse (no diagnosis), AI-DEPUR+ turns first to the ontology and then, if it fails, to the plant manager demanding an of f- line diagnosis based on their microbiological deep kno wledge. This xternal solution is learned. Detailed integration schema 1. CBRS diagnosis and RBES diagnosis:

impasse, the diagnosis inte grator turns to the ontology 2. CBRS diagnosis and RBES diagnosis: RBES diagnosis-certainty RBES diagnosis passed to de- cision support layer RBES diagnosis-certainty impasse, the diagnosis inte gra- tor turns to the ontology 3. CBRS diagnosis and RBES diagnosis: CBRS case-similarity CBRS diagnosis passed to decision support layer
Page 9
CBRS case-similarity impasse, the diagnosis inte grator turns to the ontology 4. CBRS diagnosis and RBES diagnosis: CBRS case-similarity CBRS diagnosis passed to decision support layer CBRS case-similarity RBES

diagnosis-certainty RBES diagnosis passed to de- cision support layer RBES diagnosis-certainty impasse, the diagnosis inte- grator turns to the ontology Example (case 1) ha certain perception state The case- based reasoning system (CBRS) and the rule-based xpert system (RBES) acti ate. The CBRS finds case similar to state The similarity alue is 0.1 and it is less than the minimum acceptable alue (e.g. b=0.2). Therefore there is no diagnosis output from the CBRS. The RBES finds no rules leading to diagnosis starting from state Therefore there is no diagnosis output from the RBES.

The diagnosis inte grator ackno wledges case of missing diagnosis from the KBSs and send petition to the ontology with the descrip- tion of state The output of the ontology is: Answer ’W ha foaming situ- ation’, Explanation ’I recei ed information related to the answer from the acti ation of the follo wing relation (Nocardia is Acti- nomycetes) and the follo wing axioms (Actinomycetes is cause of oaming sludge)’. Comment: State is characterized by strong presence of Nocar dia bacterium, ut this bacterium has no direct relation (in the static kno wledge-base of the RBES) to an state of the WWTP

Ev en in the ontology the Nocardia class has no link to an state of the WWTP ut its parent class (Actinomycetes) has cause-ef fect link to the gen- eral state of the WWTP ’F oaming sludge’. One reason for this could be that Nocardia is not causing foaming, ut another (not detected) bacterium of the same taxonomic class is. The diagnosis inte grator recei es the diagnosis from the ontology and pass it to the decision support layer 3.4 Decision support layer Ev entually we describe the supervisory le el of the AI-DEPUR+ en vironmental decision-support system. Once the diagnoses of the reasoners

(case-based and rule-based) and possibly of the ontology for the management of the waste water treatment ha been inte- grated, it selects an actuation. Real time This layer runs always in real time and it is ery ro- ust in this sense, simply because of the fact that the entire system minimum-c ycle time is ery long with respect to the calculations done by the decision support layer 3.4.1 Pr ediction The result of diagnosis inte gration (carried out among case-based reasoning, rule-based reasoning and the ontology) serv es as input for the prediction phase. subsystem, based on transition

networks, pre- dicts arious alternati olutions of the state of the WWTP sec- ond subsystem aluates these alternati es. The result is passed on to the actuation selector Actuation selection Actions to be carried out are selected. Often, action schemas are already included in the diagnosis result. CONCLUSIONS AND FUTURE WORK In this paper we presented the AI-DEPUR+ decision support sys- tem, inte grating rule-based xpert system, case-based reasoner and an ontological kno wledge-base. This system is able to model the information about waste water treatment process. The main im- pro ements of

AI-DEPUR+ system with respect to xistent system are: Impasse situations in xistent systems are solv ed by the ontology While in xistent systems there is no modeling of waste water mi- crobiology in this ne system the microbiological component is modeled by the ontology AI-DEPUR+ presents no el inte gration between KBSs and on- tologies in real world application. AI-DEPUR+ facilitates its wn portability AI-DEPUR+ incorporates cause-ef fect reasoning. AI-DEPUR+ will incorporate predicti skills. It will be possible semi-automatic generation of static KB. Acknowledgments This research is supported

by the Generalitat de Catalun ya (grant 1997FI 00630 UPC PG). REFERENCES [1] Alam an, X., Romero, S., Aguirre, C., Serrahima, ., Mu noz, R., opez, ., Dorronsoro, J. and de Pablo, E., 1992. MIP: Real ime Expert System in the Proceedings of the 8th Conf. On Artificial Intel- ligence Application (CAIA-92), Monterrey California. [2] ouris, N.M., 1995. Co-operating Knowledge-Based Systems for En vir onmental Decision-Support in Kno wledge-Based Systems 8(1) (1995), pp. 39-53. [3] Baeza, J., Gabriel, D. and Lafuente, J., 1998. An Expert Supervisory System for Pilot WWTP in ECAI 98 W7 (BESAI98)

workshop notes, pp. 25-37. [4] Barnett, M.W and Andre ws, J.F ., 1990. Knowledge Based Systems for Operation of astewater eatment Pr ocesses in Instrumenta- tion, Control and Automation of ater and aste water reatment and ransport Systems, Proceedings of the 5th IA WPRC workshop, oko- hama and yoto, pp. 211-18, Per gamon Press, Ne ork 1990. [5] Beccari, M., 1991. Principi dei trattamenti biologici in Canziani, R. (editor) rattamento delle acque di rifiuto pp. 99-113, Istituto per l’Ambiente, Milano, Italy 1991. [6] Borst, ., Akkermans, H. and op, J., 1997. Engineering Ontologies in

International Journal of Human-Computer Studies 46, pp. 365-406, 1997. [7] Branting, K., Hastings, J.D. and Lockwood, J.A., 1997. CARMA: pest management advisory system in AI Applications 11(1) (1997), pp. 29-48. [8] Brooks, R.A., 1986. ob ust layer ed contr ol system for mobile obot in IEEE J. Rob Aut. (RA) 2(1) (1986), pp. 14-23. [9] Buchanan, B. and Smith, R., 1988. Fundamentals of xpert systems in Annual Re vie ws on Computer Science (1988), pp. 23-58. [10] Comas, Q., R-Roda, I., Ceccaroni, L. and anchez-Marr e, M., 1999. Semi-automatic learning with quantitative and qualitative featur es

in Proceedings of the VIII Conference of the Spanish Association for Artificial Intelligence (CAEPIA-99), November 16-19, 1999, Murcia, Spain. [11] Cort es, U. and anchez-Marr e, M. (editors), 1998. Binding en vir on- mental sciences and artificial intelligence ECAI 98 W7 (BESAI98) workshop notes. [12] Damiani, A., 1991. Caratteristiche dei liquami urbani in Canziani, R. (editor) rattamento delle acque di rifiuto pp. 11-42, Istituto per l’Ambiente, Milano, Italy 1991.
Page 10
[13] Droste, R. L., 1997. Theory and practice of water and wastewater tr eatment iley 1997.

[14] Dym, C.L. and Le vitt, R.E., 1991. Knowledge-based Systems in Engi- neering McGraw-Hill 1991. [15] Farquhar A., Fikes, R., Pratt, and Rice, J., 1995. Collaborative Ontology Construction for Information Inte gration technical report KSL-95-63, August 1995, Kno wledge Systems Laboratory Depart- ment of Computer Science, Stanford Uni versity Stanford, US. [16] Farquhar A., Fikes, R. and Rice, J., 1996. The Ontolingua server: tool for collaborative ontology construction technical report KSL- 96-26, Kno wledge Systems Laboratory Stanford Uni versity Stanford, US. [17] Fedra, K., 1993. Expert

systems in water esour ces simulation and optimization in Marco, J.B. et al. (editors) Stochastic hydr ology and its use in water esour ces systems simulation and optimization pp. 397-412, Kluwer Academic Publishers, The Netherlands 1993. [18] Fedra, K., 1995. Decision support for natural esour ces managemen t: models, GIS and xpert systems in AI Applications 9(3) (1995), pp. 3-19. [19] Fikes, R., Farquhar A. and Rice, J., 1997. ools for Assembling Modu- lar Ontologies in Ontolingua technical report KSL-97-03, Kno wledge Systems Laboratory Stanford Uni versity Stanford, US. [20] Finch, .E.,

Oyeleye, O.O. and Kramer M.A., 1990. Rob ust Event- Oriented Methodology for Diagnosis of Dynamic Pr ocess Systems in Computers Chemical Engineering 14(12) (1990), pp. 1379-96. [21] omez-P erez, A., 1998. Knowledge sharing and euse in the Hand- book of Applied Expert Systems, pp. 10.1-10.36, CRC Press 1998. [22] omez-P erez, A., Fern andez, M. and de icente, A., 1996. owar ds method to conceptualize domain ontologies in the proceedings of the workshop on Ontological Engineering, ECAI’96, Budapest, Hungary pp. 41-52. [23] Gonz alez and Dankel, 1994. The engineering of knowledge-ba sed sys- tems

Prentice-Hall 1994. [24] Gruber .R., 1993. ranslation Appr oach to ortable Ontologies in Kno wledge Acquisition 5(2), pp. 199-220, 1993. [25] Guariso, G. and erthner H. (editors), 1989. En vir onmental Decision Support Systems Ellis Horwood iley 1989. [26] Jackson, ., 1990. Intr oduction to xpert systems Addison esley 1990. [27] Jones, D.M., Bench-Capon, .J.M. and isser .R.S., 1998. Method- ologies for Ontology Development IT&KNO WS Information echnology and Kno wledge Systems, 15th IFIP orld Computer Congress, ienna (Austria) and Budapest (Bulgaria). [28] Kaelbling, L.P ., 1990. An ar

chitectur for intelligent eactive systems in Allen, J., Hendler J. and ate, A. (editors) Readings in Planning pp. 713-28, Mor gan Kaufmann 1990. [29] Krovvidy S. and ee, .G., 1991. astewater tr eatment systems fr om case-based easoning in Machine Learning 10 (1991), pp. 341-63. [30] Mason, C. (editor), 1995. orkshop on Artificial Intelligence and the En vir onment in IJCAI-95 orkshop Program orking Notes. [31] Matwin, S., Charlebois, D., Goodenough, D.G. and Bhogal, ., 1995. Machine learning and planning for data management in for estry in IEEE Expert Systems 10 (5) (1995). [32] Neches,

R., Fikes, R., Finin, ., Gruber ., Patil, R., Senator and Swartout, .R., 1991. Enabling echnology for Knowledge Sharing in AI Magazine (winter 1991), pp. 36-56. [33] Nov otny ., Jones, H., Feng, X. and Capodaglio, A.G., 1990. ime Series Analysis Models of Activated Sludge Plants in ater Science echnology 23(4-6) (1990), pp. 1107-16. [34] Ortolano, L., LeCoeur G. and MacGilchrist, R., 1990. Expert Sys- tem for Sewer Network Maintenance: alidation Issues in Journal of Computing in Ci vil Engineering 4(1) (1990), pp. 37-54. [35] Patry G. and Chapman, D. (editors), 1989. Dynamic modelling and

xpert systems in wastewater engineering Le wis Publisher Chelsea, MI, USA 1989. [36] anchez-Marr e, M., 1995. AI-DEPUR: an inte grated supervisory multi-level ar chitectur for wastewater tr eatment plants Ph.D. The- sis, Software Department, UPC, Barcelona, Spain. [37] anchez-Marr e, M., Cort es, U., R-Roda, I. and Poch, M., 1998. Case Learning thr ough Similarity Measur for continuous domains in ECAI 98 W7 (BESAI98) workshop notes, pp. 39-53. [38] Sanz, R., Aguilar J., Sierra, C., God o, L. and Ollero, A., 1989. Adap- tive contr ol with supervisor level using rule-based infer ence sys- tem

with appr oximate easoning in Artificial Intelligence in Scientific Computation: to wards 2nd generation systems, IMA CS, 1989. [39] Serra, ., 1993. Development of knowledge-based system for contr ol and supervision of urban wastewater tr eatment plants Ph.D. Thesis, Chemistry Department, AB, Barcelona, Spain. [40] Shepherd, A. and Ortolano, L., 1996. ater -supply system operations: Critiquing xpert-system appr oach in J. Of ater Resources Planning and Management 122(5) (1996), pp. 348-55. [41] Stephanopoulos, G., 1990. Artificial intelligence in pr ocess engineer ing: curr

ent state and futur tr ends in Computers Chemical Engi- neering 14 (1990), pp. 1259-70. [42] Wright, J., iggins, L., Jain, R. and John Kim, (editors), 1993. Expert systems in En vir onmental Planning Springer -V erlag, Berlin- Heidelber g-Ne wY ork 1993. [43] ang, C.T and Kao, J.J., 1996. An xpert-system for selecting and se- quencing wastewater tr eatment pr ocesses in ater Science ech- nology 34(3-4) (1996), pp. 347-53. [44] Zhu, X.X. and Simpson, A.R., 1996. Expert System for ater eat- ment Plant Operation in Journal of En vironmental Engineering 122(9) (1996), pp. 822-29. 10