Optimizing Operational Policies of a Korean Multireservoir System Using Sampling Stochastic Dynamic Programming with Ensemble Streamflow Prediction

Abstract

This study presents state-of-the-art optimization techniques for enhancing reservoir operations which use sampling stochastic dynamic programming (SSDP) with ensemble streamflow prediction (ESP). SSDP used with historical inflow scenarios (SSDP/Hist) derives an off-line optimal operating policy through a backward-moving solution procedure. In contrast, SSDP used with monthly forecasts of ESP (SSSDP/ESP) reoptimizes the off-line policy. These stochastic models are used to derive a monthly joint operating policy during the drawdown period of the Geum River multireservoir system in Korea. A cross-validation test of 1,900 simulation runs demonstrates that: (1) proposed stochastic models that explicitly include inflow uncertainty are superior to those that do not; (2) updating policy with ESP forecasts is appropriate in this reservoir system; (3) the lower dam of the Geum River multireservoir system should maintain elevation of 66.5m during the beginning of the drawdown period to avoid significant increase in the downstream water shortages; and (4) forecasting accuracy may result in considerable effects on joint reservoir operations.