Ensemble of constraint-handling techniques for solving reservoir scheduling problems

Abstract

Abstract Reservoir scheduling based on evolutionary algorithms needs to handle potentially stringent physical and operational constraints. Both generic and reservoir scheduling problem-specific constraint-handling techniques (CHTs) have their own merits and limitations. No CHT currently available can yield better solutions than the others consistently. To ensure good reservoir operation schedules, we develop an ensemble of CHTs (ECHT) that can utilize the advantages of different individual CHTs. In the ensemble, each CHT has its own population. In every generation, the different offspring populations are mixed together and evaluated. Each CHT then assigns fitness to all individuals and selects some of them to form its new parent population. The ECHT has been tested against long-term hydropower scheduling of two large-scale reservoir systems in China. Results show that the ECHT outperforms the state-of-the-art CHTs, and its probability of returning feasible solutions is much higher. The reservoir levels optimized with the ECHT are well suited for hydropower generation, which also reduce the chance of reservoir spilling.