適応的探索による火力発電プラント起動スケジューリング

抄録

Start-up scheduling is aimed at minimizing the start-up time while limiting turbine rotor stresses to an acceptable level. This scheduling problem has a wide search space. In order to improve the search efficiency and robustness and to establish an adaptive search model, we propose to integrate evolutionary computation, based on Genetic Algorithms (GA), with reinforcement learning. The strategies with our proposal include: multi-boundary-based enforcement operator and multi-elitist plan. By setting a second boundary, located right outside the existing boundary containing those feasible schedules, we extend our proposed enforcement operator and the conventional elitist plan into the multi-boundary-based enforcement operator and multi-elitist plan. These two strategies work together to focus the search along the boundary, around which the optimal schedule is supposed to exist, so as to increase the search efficiency as well as its robustness. During a search process, GA guides the reinforcement learning to concentrate its learning on those promising areas instead of the entire space. In return, reinforcement learning can generate a good schedule, in the earlier stage of the search process. We obtain encouraging test results. In this paper, we propose the GA-based search model with these strategies and discuss the test results.