Reinforcement learning-based modified cuckoo search algorithm for economic dispatch problems

Abstract

One of the crucial tasks of social development is carbon neutrality, which is mainly due to the emission of greenhouse gases. However, thermal power, which produces plenty of harmful gases, is still the main component of electric energy. Therefore, Economic Dispatch (ED) is proposed to utilize energy resources more efficiently and reduce the cost of power generation. ED is a nonlinear and nonconvex-constrained optimization problem that is difficult to optimize. In this paper, we propose a Reinforcement Learning-based Modified Cuckoo Search algorithm (RLMCS) to solve ED problems. The proposed algorithm employs the concept of Reinforcement Learning (RL) and develops an RL-based method to process population obtained from the explorative phase. The RL-based method can dynamically enhance the population based on cumulative rewards and the current environmental state. Thus, the comprehensive search ability of RLMCS has been well improved. Moreover, some proven technologies, i.e., Gaussian random walk, quasi-opposition learning, and adaptive switch parameter, are introduced to further enhance the efficiency of RLMCS. The performance of the RLMCS is tested on standard ED problems (6 and 11 units) and ED problems with valve-point effects (10, 14, and 40 units). RLMCS is also compared with some well-established CS variants. The experimental results have demonstrated that RLMCS is more competitive and robust.