Multi reservoir systems optimisation using genetic algorithms

The operation of multiple reservoir systems for sometimes conflicting purposes can be a complex process because of the involvement of a large number of decision variables and constraints. Dynamic programming (DP) has long been recognised as a powerful approach in the analysis of water resource systems. The usefulness of DP for multi reservoir systems is, however, limited by the huge demand that it can induce on computational resources. Many forms of DP have been developed to alleviate the problem of dimensionality with varying degrees of success, but no general algorithm exists. This thesis describes the development and application of genetic algorithms (GAs) for the optimisation of multi reservoir systems. The GA approach is validated through application to a number of problems with known solutions. Several alternative formulations of a GA for reservoir systems are evaluated using the four reservoir problem. This has been done with a view to presenting fundamental guidelines for implementation of the approach to practical problems. Alternative representation, selection, crossover, and mutation schemes are considered. The most promising GA approach comprises real-value coding, tournament selection, uniform crossover, and modified uniform mutation. A non-linear four reservoir problem was also solved, along with a problem with extended time horizons. A more complex ten reservoir problem was also successfully solved. The practicality of the developed GA approach in the determination of optimal reservoir operating rules is considered through application to a reservoir system in Indonesia. Optimal operating rules have been derived for the existing development situation in the basin, and for two future water resource development scenarios, using critical period hydrology. A comparison of the GA results with those produced by discrete differential DP (DDDP) is also presented. The application of GA approach to real time operations with stochastically generated inflows is also demonstrated for the Equatorial Lakes system in Africa. A methodology for forecasting reliable power that can be produced over different durations of time has also been developed using a GA. For the problems considered in this study, the GA solutions are very close to the optimum. The results demonstrate that the approach is robust and is easily applied to complex systems. It has potential as an alternative to stochastic DP approaches.