Evaluation of Genetic Algorithms for Optimal Reservoir System Operation

Abstract

Several alternative formulations of a genetic algorithm for reservoir systems are evaluated using the four-reservoir, deterministic, finite-horizon 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. It is concluded that the most promising genetic algorithm approach for the four-reservoir problem comprises real-value coding, tournament selection, uniform crossover, and modified uniform mutation. The real-value coding operates significantly faster than binary coding and produces better results. The known global optimum for the four-reservoir problem can be achieved with real-value coding. A nonlinear four-reservoir problem is considered also, along with one with extended time horizons. The results demonstrate that a genetic algorithm could be satisfactorily used in real time operations with stochastically generated inflows. A more complex ten-reservoir problem is also considered, and results produced by a genetic algorithm are compared with previously published results. The genetic algorithm approach is robust and is easily applied to complex systems. It has potential as an alternative to stochastic dynamic programming approaches.