Hybrid Two-Stage Stochastic Methods Using Scenario-Based Forecasts for Reservoir Refill Operations

Abstract

Optimizing reservoir refill operations is important for improved water use. This study developed a real-time refill operation model using scenario-based forecasts. To bridge the gap between forecast horizon and operation horizon, the future operation horizon was divided into two stages based on the forecast horizon point. The first stage (the forecast horizon) was provided with scenario-based forecasts, while the second stage (the remaining horizon) was described using historical streamflow scenarios. Based on deterministic dynamic programming (DDP), explicit stochastic optimization (ESO), and implicit stochastic optimization (ISO), three hybrid two-stage stochastic methods (ESO-DDP, ISO-ESO, and ISO-DDP) were proposed. Using China’s Three Gorges Reservoir as a case study, the performances of six schemes (DDP, stochastic dynamic programming, sampling stochastic dynamic programming, ESO-DDP, ISO-ESO, and ISO-DDP) were compared. Results showed that ISO-DDP performed best in terms of refill rate and hydropower generation. Specifically, the ISO-DDP scheme decreased the refill rate by 2.33% and decreased the hydropower generation by 2.31% compared with the DDP scheme. Therefore, the proposed ISO-DDP method could be useful in reservoir refill operations.