Reinforcement Learning-Based Differential Evolution for Solving Economic Dispatch Problems

In power systems, economic dispatch (ED) deals with the power allocation of power generation units to meet the power demand and minimize the cost. Many metaheuristics have been proposed to solve the ED problem with promising results. However, the performance of these algorithms might be sensitive to their parameter settings, and parameter tuning requires considerable effort. In this paper, a reinforcement learning (RL)-based differential evolution (DE) is proposed to solve the ED problem. We develop an RL mechanism to adaptively set two critical parameters, crossover rate (CR) and scaling factor (F), of DE. The performance of the proposed RLDE is compared with the canonical DE and several algorithms in the literature using three test systems. Our algorithm shows good solution quality and strong robustness.