Memetic reinforcement learning based maximum power point tracking design for PV systems under partial shading condition

摘要

Solar energy has attracted significant attentions around the globe, while one of its most crucial task is to harvest the maximum available solar power under different weather conditions, also known as maximum power point tracking (MPPT). This paper proposes a novel memetic reinforcement learning (MRL) based MPPT scheme for photovoltaic (PV) systems under partial shading condition (PSC). In order to enhance the searching ability of MRL, the memetic computing structure is incorporated into reinforcement learning (RL). In particular, a virtual population is used for the global information exchange between different agents, such that the learning rate can be dramatically accelerated. Besides, a RL based local search is designed in each memeplex, which can effectively improve the optimum quality. Comprehensive case studies are undertaken, such as start-up test, step change of solar irradiation, ramp change of solar irradiation and temperature, and field atmospheric data of Hong Kong. The PV system responses are then evaluated and compared to that of seven typical MPPT algorithms. (C) 2019 Elsevier Ltd. All rights reserved.