SSDP Model with Inflow Clustering for Hydropower System Operation

Abstract

Abstract Sampling stochastic dynamic programming (SSDP), which considers the uncertainty of streamflow, is a popular and useful method for solving release decisions of reservoirs. It is easy to implement the long-term operation for cascaded hydropower stations with poor inflow prediction ability. Furthermore, SSDP describes the empirical distribution by inflow scenarios considering the temporal and spatial structure of the streamflow processes instead of dividing the inflow into discrete representative values in stochastic dynamic programming (SDP). However, the computation time of the procedure will increase exponentially with the growth of reservoirs and inflow scenarios. Thus, the clustering method is employed to reduce the inflow scenarios in order to improve the efficiency and operability of SSDP in practical applications. The calculation results and the improvement on computation time consumption are analyzed with different cluster numbers in clustering algorithm. The principle of how to select the least inflow scenarios to represent all inflow sequences has also been proposed. Results show that the SSDP model with clustered inflows scenarios can significantly reduce the computation time. The least inflow scenarios selected by clustering algorithm can represent the empirical distribution of 56 streamflow scenarios without obviously decreasing energy and exacerbating the shortage of firm power in results in this study.