Optimizing urban water supply headworks using probabilistic search methods

Simulation models in conjunction with synthetic multiple hydro-climate replicates provide the most realistic assessment of the performance of urban water supply headworks systems. However, optimization using such models is computationally very demanding. To appreciate the challenge, a simple case stud system is presented. The system consists of one reservoir, has three decision-variables to be optimized, and uses an objective function based on reservoir costs and economic penalties for water shortages. Enumeration revealed that the objective function surface has piecewise flat regions that arise from operating rule thresholds and the infrequent sampling of severe droughts. Two search methods capable of dealing with such flat regions, the genetic algorithms (GA) and the shuffled complex evolution (SCE) method, were investigated. For the GA to be robust (i.e., avoid premature convergence on flat regions), it was necessary to employ two lesser-known genetic operators, inversion and population selection strategy. The SCE method was found to have comparable robustness, but required fewer evaluations. Nonetheless, the GA method was preferred because of its inherent advantage in parallel computing.