Reinforcement Learning-Based Adaptive Optimal Fuzzy MPPT Control for Variable Speed Wind Turbine

A reinforcement learning-based adaptive optimal fuzzy controller is proposed for maximum power point tracking (MPPT) control of a variable-speed permanent magnet synchronous generator-based wind energy generation system. The algorithm consists of a critic, an adaptive optimal fuzzy controller, and an adaptive optimal fuzzy estimator. The critic is built based on an adaptive neuro-fuzzy inference system (ANFIS) network instead of the neural network as normal to reduce the computation. The error between the system output and the estimator output is used as the input of the critic. In addition, the critic is used to calculate the update law for the parameters of the adaptive optimal fuzzy controller and adaptive optimal fuzzy estimator based on minimizing the input error function. Moreover, the proposed control scheme is output feedback instead of state feedback, which does not require a system model as well as system parameters, so the system is robust to uncertainties and external disturbances. Besides, the stabilization proof is accomplished by using the Lyapunov stability theorem for the closed-loop system and the convergence of the update law. Finally, the effectiveness of the proposed reinforcement learning-based adaptive optimal fuzzy control scheme is verified through simulation with various scenarios such as step wind speed, random wind speed, and system parameter variations. Also, the comparisons with other control schemes in the state-of-art (neural network reinforcement learning based adaptive optimal fuzzy controller, PI controller) are executed to demonstrate the advantages of the proposed control scheme.