Application of Dynamic Programming for the Analysis of Complex Water Resources Systems: A Case Study on the Mahaweli River Basin Development in Sri Lanka Ontvangen UQ-CARDEX 0000 0489 BibJUOlHJbUv EA£JPBOUWUNIVERSHEEt Promotoren: Dr.-Ing. J.J. Bogardi Hoogleraar in de Hydrologie, Hydraulica en Kwantitatief Waterbeheer Dr. P. van Beek Hoogleraar in de Operationele Analyse M.D.U.P. Kularathna Application of Dynamic Programming for the Analysis of Complex Water Resources Systems: A Case Study on the Mahaweli River Basin Development in Sri Lanka Proefschrift ter verkrijging van de graad van doctor in de landbouw- en milieuwetenschappen op gezag van de rector magnificus, dr. H.C. van der Plas, in het openbaar te verdedigen op vrijdag 12 juni 1992 des namiddags te half twee in de Aula van de Landbouwuniversiteit te Wageningen Tom y parents ^O2TX>I \siH Statements 1. Due toth euncertai n conditions under which theactua l operation of water resources systems takesplace flexibility, o f operation is an important aspect which should not be left aside in any operational study. This thesis. 2. Methodology for formulating long-term operation policies of water resources systems need more attention than their short-term counterparts. This thesis. 3. In contrast to the optimization models, simulation models can best be employed to assess the performance of a system if the operation policies have been predetermined. This would permit adetaile d investigation of theresultin g operation pattern and subsequent improvements toth eoperatio n policies which are derivedb y optimization models thatinclud e much less details of thesyste m than the simulation models. This thesis. 4. "Objectives" are often noncommensurate, even in a simple one-man enterprise. 5. Irrigation in Sri Lanka is a long-practised art using the traditional tank-irrigation systems. This vast experience should beintegrate d in the future development of Sri Lankas water resources systems. 6. An approximate answer to the right question is better than the right answer to a wrong question. 7. To effectively contribute to research, one should not only know how much one knows, but also how much one has yet to learn. 8. Technology must always be appropriate to the country's needs and circumstances. There is agrea t need for development of such appropriate technology, and also for guidance on ways of selecting the most cost-effective mixture of conventional and new technology. WMO/UNESCO (199D. Report on Water Resources Assessment, p. 47. 9. The communication and cooperation between the universities and the practitioners need to be improved, in order to narrow the existing gap between theory and practice in the field of water resources management. W S / 10. In all scientific research, theresearche r may or mayno t find what he/shei slookin g for. Indeed his/her hypothesis may be demolished. But he/she is certain to learn something, which may be and often is more important than what he/she had hoped to learn. Robert Heinlein, in: Richard Boyle (1991), The Serendipity Factor, Serendib (The Magazine of Air Lanka1). Vol. 10, No. 4. 11. The environmental impacts of large scale water resources developments havet o be carefully investigated in theplannin g stage, as thoseimpact sar e often irreversible. 12. The Forest, With endless life-giving qualities, It protects all living beings And provides shelter Even to those who destroy it with an axe. The Lord Buddha. M.D.U.P. Kularathna Application of Dynamic Programming for the Analysis of Complex Water Resources Systems: A Case Study on the Mahaweli River Basin Development in Sri Lanka Wageningen, 12* June 1992 Abstract Kularathna, M.D.U.P. (1992), Application of Dynamic Programming for the Analysis of Complex Water Resources Systems: A Case Study on the Mahaweli River Basin Development in Sri Lanka, Doctoral Dissertation, Wageningen Agricultural University, Wageningen, The Netherlands, (xviii) + 163 pp., 32 Figures, 32 Tables (Summary and Conclusions in English and Dutch) The technique of Stochastic Dynamic Programming (SDP) is ideally suited for operation policy analyses of water resources systems. However SDP has a major drawback which is appropriately termed as its "curse of dimensionality". Aggregation/Disaggregation techniques based on SDP and simulation are presented to analyze a complex water resources system. The system under consideration serves two major purposes: hydropower generation and irrigation. Theidentificatio n of subsystemsb y their functional and physical characteristics was an important first step in the analysis. Subsequently each subsystem isrepresente d by ahypothetica l compositereservoi r toarriv e at an operation policy for the interface point of the subsystems. A more detailed analysis which considers the real configurations of the subsystems is performed by following this operation policy of the interface point. Two approaches: sequential optimization and iterative optimization are presented. In these approaches, each subsystem is individually analyzed using two-reservoir SDP models. The applicability of an Implicit Stochastic Approach in which the operation of the system is optimized for a number of deterministic hydrologic data series is also investigated. To complement the aggregation technique of the Composite Reservoir, subsequent disaggregation techniques are proposed. Three different techniques: (1) A statistical disaggregation, (2)A noptimization/simulation-base d technique,an d(3 )Th edisaggregatio n of the composite policy in the actual operation by incorporating a single-time-step optimization are tested. The accuracy of the sequential and iterative optimization approaches are evaluated by applying them to a subsystem of three reservoirs in a cascade for which the deterministic optimumpatter n isals odetermine d bya nIncrementa lDynami cProgrammin g (IDP)model . In the case of the Implicit Stochastic Approach, the results are compared with the results of theexplici t SDPapproac h and the deterministic optimum operation pattern, in addition to the historical operation pattern of the system. The results of the Composite Policy Disaggregation techniques are compared to the results obtained by real multireservoir optimizations carried out by the use of explicit SDP models. Samenvatting Kularathna, M.D.U.P. (1992), Application of Dynamic Programming for the Analysis of Complex Water Resources Systems: A Case Study on the Mahaweli River Basin Development in Sri Lanka, proefschrift, Landbouwuniversiteit Wageningen, Wageningen, Nederland, (xviii) + 163 pp., 32 figuren, 32 tabellen. (Samenvatting en Conclusies in Engels en Nederlands) De techniek Stochastic Dynamic Programming (SDP) is zeer geschikt voor analyse van beheer van water systemen. SDP heeft echter een belangrijk knelpunt, welke wordt aangeduid met haar 'curse of dimensionality'. Aggregatie/disaggregatie technieken, gebaseerd op SDP, worden gepresenteerd ten einde complexe water systemen te analyseren. Het beschouwde systeem heeft twee belangrijke functies :waterkrach t opwekking en irrigatie. Een eerstebelangrijk e stapi n deanalys ewa s de identificatie van subsystemen op grand van hun functionele en fysische eigenschappen. Vervolgens werd ieder subsysteem gemodelleerd door een hypothetisch samengesteld reservoir, omz o totee n optimaalbehee r voor deverbindingspunte n tussen de subsystemen te komen. Daarna werd een meer gedetailleerde analyse uitgevoerd, uitgaande van de werkelijke configuratie van de subsystemen en het eerder gevonden beheer op de verbindingspunten. Twee benaderingen worden beschreven: sequentiele en iteratieve optimalisatie. In deze benaderingen wordt ieder subsysteem individueel geanalyseerd als SDP modellen bestaande uit twee reservoirs. De toepasbaarheid van een impliciete stochastische benadering, waarbij het systeem geoptimaliseerd wordt voor een aantal deterministische hydrologische reeksen, is ook onderzocht. Ter aanvulling van de aggregatie techniek van het samengestelde reservoir worden disaggregatie technieken voorgesteld. Drie technieken worden getest: (1) een statistische disaggregatie techniek, (2) een techniek gebaseerd opoptimalisatie/simulati ee n (3) disaggregatie van het huidige, samengestelde beleid door optimalisatie per enkele tijdstap. De nauwkeurigheid van de sequentiele en iteratieve optimalisatie methoden worden geevalueerd door hen toe tepasse n op een systeem van drie reservoirs in serie, waarvoor ookme tbehul pva nIncrementa l DynamicProgrammin g (IDP)he tdeterministisch eoptimal e beheer werd bepaald. De resultaten van de impliciete stochastische methode worden vergeleken met de resultaten van de expliciete SDP methode, met het deterministisch optimale beheer, en met het historische beheer van het systeem. De resultaten van de disaggregatie technieken voor samengesteld beheer wordenvergeleke n metd eresultate n die verkregen door metbehul pva n SDPee n systeem van meerderereservoir s teoptimaliseren . VI Acknowledgements The author wishes to express his gratitude to Prof. Dr.-Ing. J.J. Bogardi and Prof. Dr. P. vanBee k for giving himth eopportunit y tocarr y outth epresen t study under their guidance. Their constructive criticisms and continued encouragements during the courseo fthi sstud y wereinvaluable .Th e helpgive n inman yway sb yProf . Bogardi during thelas t fiveyear si sgratefull y acknowledged, although theautho r feels thati ti s insufficient to express hisgratitud e in a few words or sentences. It iswit h gratitude theautho r acknowledges theguidanc e given byProf . R. Harboe atth e Asian