Optimal mileage-based PV array reconfiguration using swarm reinforcement learning

• Published in: Energy conversion and management Vol. 232; p. 113892
• Main Authors: Zhang, Xiaoshun, Li, Chuanzhi, Li, Zilin, Yin, Xueqiu, Yang, Bo, Gan, Lingxiao, Yu, Tao
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 15.03.2021; Elsevier Science Ltd
• Subjects: administrative management; Algorithms; Arrays; butterflies; Electric power distribution; Electric power generation; energy conversion; Genetic algorithms; grasshoppers; Learning; Machine learning; markets; Optimization algorithms; Partial shading condition; Particle swarm optimization; Power plants; PV array reconfiguration; Real-time generation scheduling; Reconfiguration; Regulation mileage; Reinforcement; Shading; solar farms; Swarm reinforcement learning; swarms; Regulation mileage; Partial shading condition; PV array reconfiguration; Swarm reinforcement learning; Real-time generation scheduling
• ISSN: 0196-8904; 1879-2227
• DOI: 10.1016/j.enconman.2021.113892

•A new optimal mileage-based PV array reconfiguration (OMAR) is constructed.•The OMAR can maximize the total benefit instead of only the generation benefit.•The OMAR decomposition with two sub-problems reduces the optimization difficulty.•The swarm reinforcement learning is used to obtain high-quality optimums of OMAR.•The proposed method can obtain higher total benefit than 6 comparative algorithms. This paper constructs a new optimal mileage-based PV array reconfiguration (OMAR) in a PV power plant under partial shading conditions. It aims to maximize the power output of a PV power plant, and minimize the additional capacity and mileage payments resulting from the power fluctuation in a performance-based frequency regulation market. To reduce the optimization difficulty of OMAR, it is decomposed into two optimization sub-problems, including an upper-layer discrete optimization of PV array reconfiguration and a lower-layer continuous optimization of real-time generation scheduling. The upper-layer discrete optimization is addressed by the proposed swarm reinforcement learning (SRL), which can implement an efficient exploration and exploitation with multiple cooperative agents instead of a single learning agent. The rest lower-layer optimization is handled by the fast interior point method. The proposed method’s effectiveness is thoroughly evaluated on the 10 × 10 total-cross-tied PV arrays under various partial shading conditions. Simulation results demonstrate that the proposed SRL can obtain a larger total benefit than genetic algorithm (GA), particle swarm optimization (PSO), grasshopper optimization algorithm (GOA), harris hawks optimizer (HHO), butterfly optimization algorithm (BOA), and Q-learning, in which the benefit increment can reach from 2.12% (against PSO) to 10.62% (against Q-learning).