Sampling stochastic dynamic programming applied to reservoir operation

Abstract

Most models for reservoir operation optimization have employed either deterministic optimization or stochastic dynamic programming algorithms. This paper develops sampling stochastic dynamic programming (SSDP), a technique that captures the complex temporal and spatial structure of the streamflow process by using a large number of sample streamflow sequences. The best inflow forecast can be included as a hydrologic state variable to improve the reservoir operating policy. A case study using the hydroelectric system on the North Fork of the Feather River in California illustrates the SSDP approach and its performance.