Efficient Tabu Search Procedure for Short-Term Planning of Large-Scale Hydropower Systems

Abstract

Short-term hydrogeneration scheduling aims at minimizing water consumption for the next 7–15 days on an hourly basis, while satisfying the electrical load as well as many operational, regulatory, and safety requirements. In an ever-changing environment, planners need to make decisions quickly and often adapt their schedules to new conditions. They need a tool that is fast, reactive, and flexible. This paper presents a new solution approach that provides, within a few minutes of computation, near-optimal solutions to hard problems, in which one seeks to determine the number of committed generating units and turbined and spilled flows on an hourly basis for a planning horizon of 10 days. Our solution approach is based on tabu search with new neighborhoods, allowing simultaneous modifications for several plants and time periods. It can handle multiobjective problems, as well as nonlinear and nonconvex constraints. A decomposition mechanism allows parallelism and search acceleration for long horizons and large production systems. Computational experiments on real instances from Hydro-Québec show that even with the most naïve initial solution, our approach finds solutions almost as good as mixed integer linear programming solutions up to 235 times faster than conventional solvers and with less modeling limitations. Starting from a previous planning solution yields even better results.