Maximising water supply system yield subject to multiple reliability constraints via simulation-optimisation

Abstract

The realistic incorporation of reliability into the optimisation of reservoir system design and operation remains a particularly difficult task after decades of research. While most of this research has worked with methods based on linear or dynamic programming, little has been done to find out how well the problem could be handled by a simulation model linked to an optimisation model (SO model). Water supply systems have to satisfy different demands that each require various levels of reliability and these need to be incorporated in analyses for efficient system design and operation. This study presents an approach for determining the reservoir sizes and monthly operating rules that maximise the yield of a water supply system subject to multiple reliability constraints of supply and reservoir storage. A behaviour analysis model linked to a genetic algorithm is applied and the constraints are implemented using multiplicative penalties. This approach is found to deal with multiple reliability constraints realistically and effectively with multiple runs clearly identifying the active and the redundant constraints. The long computation times are, however, a drawback for the approach and suggestions to reduce these are suggested. Powell's conjugate direction method is also used to optimise one of the cases analysed and obtains a slightly lower yield than the genetic algorithm but with a lower number of simulations. The method obtains yields comparable to the South African Water Resources Yield Model (WRYM) and has the advantage of automating the derivation of inter-reservoir operating rules.