Multi-Agent Reinforcement Learning-Based Maximum Power Point Tracking Approach to Fortify PMSG-Based WECSs

Wind energy conversion systems (WECS) adopting conventional maximum power point tracking (MPPT) approaches, may experience issues including slow tracking speeds, poor precision, and hypersensitivity to fluctuations in wind speed. Present study develops a multi-agent reinforcement learning (MARL) strategy to overcome the limitations of conventional MPPT in variable wind speed WECS. Compare to the traditional methods, the proposed MARL approach, promote energy output, and the capability of responding swiftly to variation in wind speed. Moreover, owing to the decentralized nature of the method, several agents as opposed to a single one, would cooperate together to maximize power generation. This would guarantee superior precision, and enhanced interaction capacity over single agent reinforcement learning (RL) based methods. Also, by involving meta-learnt discount factor (DF), the advised MARL algorithm is further enhanced in terms of learning phase time, and convergence rate, leading to a more robust solution. Extensive simulation results are provided to offer comprehensive performance assessment of the proposed model-free MARL approach, highlighting its potential applicability. Moreover, a 1000 W prototype is also implemented to verify the functionality of the proposed algorithm for MPPT scheme under real-world wind conditions.