Optimal Selective Withdrawal Rules Using a Coupled Data Mining Model and Genetic Algorithm

Abstract

This work presents a methodology for extracting optimal operational rules for selective reservoir water withdrawal by considering fixed levels of reservoir water outlets for thermal control. The outlet water temperature of the Karkheh reservoir, Iran, is simulated with the CE-QUAL-W2 model. A data-mining model (the LIBSVM model) is applied as a surrogate model of the CE-QUAL-W2 model and coupled with a genetic algorithm (GA), resulting in the LIBSVM-GA algorithm. The selective withdrawal approach considered four fixed reservoir outlets, located at 120, 140, 163, and 181 m above sea level, to account for reservoir thermal stratification. This paper’s methods are evaluated with nonselective and selective withdrawal operations through different scenarios in which single outlet, fixed withdrawal proportions, fixed monthly variable proportions, continually variable (10-day) proportions using total monthly LIBSVM input data, and continually variable (10-day) proportions using separated monthly LIBSVM input data are considered. The highest outlet (at 181 m) was found to be the best level for the nonselective withdrawal scenario. The best selective withdrawal operations scenario was the continually variable (10-day) proportions using separated monthly LIBSVM input data, which minimize the root-mean-square deviation (RMSD) between upstream and downstream temperatures during the operating period.