Distributed Parallel Stochastic Dynamic Programming Algorithm Based on MPI and Multithread for Long-Term Optimal Operation of Cascaded Hydropower Stations

Stochastic dynamic programming (SDP) is widely adopted in a long-term optimal operation of large-scale hydropower systems. In this paper we propose a distributed parallel stochastic dynamic programming algorithm based on Message Passing Interface (MPI) and a peer-to-peer parallel paradigm (DPSDPoM). To deal with the disadvantages of redundancy in communications and memory-consumption during calculation in the peer-to-peer parallel paradigm, we propose a DPSDPoM with multithread algorithm (DPSDPoM-MT) which reduces costs between processes in each machine. The two algorithms are compared through the optimization scheduling of three reservoirs on Lancang Jiang Dam Cascade using time-elapse and memory consumption. Experimental results demonstrate that the improved algorithm can reduce computing time and alleviate memory consumption effectively.