Short-term hydrothermal generation scheduling using a parallelized stochastic mixed-integer linear programming algorithm

Short-term hydrothermal generation scheduling (STHTGS) is the optimization process through which decisions are made about the commitments of thermal generators and the allocation of hydro energy resources in the planning horizon (1 day to 1 week), while satisfying a large set of technical constraints. Uncertainty in this problem may appear in different modelling parameters, but the extended stochastic version of the STHTGS problem may lead to impractical solution times. This paper discusses the application of a parallelized stochastic mixed-integer linear program (SMILP) to solve the stochastic STHTGS problem. In order to decrease simulation time a scenario-based decomposition approach based on the progressive hedging (PH) algorithm is proposed. Computational experiments are conducted in two multi-processor nodes of a cluster for different numbers of stochastic scenarios. The algorithm is tested in the Chilean Central Interconnected System using a problem instance considering a weekly horizon with hourly resolution. Results show that the PH algorithm has good convergence properties, needing only a few iterations to converge. Furthermore, as PH generates similarly sized sub-problems, the parallel version of the algorithm scales up quite well as the number of scenarios is increased. (C) 2016 The Authors. Published by Elsevier Ltd.