Deep Reinforcement Learning-Based Control Scheme for Performance Enhancement of PMSG Wind Turbine With Vienna Rectifier

A novel control scheme based on deep reinforcement learning (DRL) is presented to improve the operational performance of a permanent magnet synchronous generator (PMSG) with a Vienna rectifier (PGVR) in a wind turbine generator system. This article investigates the harmonics problem of stator current in PMSG and the fluctuation problem of neutral point voltage (NPV) in the Vienna rectifier. First, a wind speed-based reward function with variable weight coefficients is designed to realize the intelligent operation of the PGVR control system. Second, a fast response Agent model with wind speed as the first observation state is established to minimize the influence of the external environment on the PGVR control system. Finally, diverse stochastic training environments are elaborated to ensure that the PGVR control system has enough experience to cope with different wind speed variation scenarios. The twin delay deep deterministic policy gradient (TD3) algorithm is used for offline training. Simulation and experimental results show that the proposed scheme has small control errors at different wind speeds and effectively improves power quality and generation efficiency.