Hybrid Fuzzy PID with Soft Actor-Critic Reinforcement Learning for Wind Turbine Pitch Control

This paper proposes a novel hybrid intelligent control framework integrating a fuzzy supervisory Proportional-Integral-Derivative (PID) controller with Soft Actor-Critic (SAC) reinforcement learning for pitch angle regulation in variable-speed wind turbines. The fuzzy supervisory layer employs seven triangular membership functions for both rotor speed error and its rate of change, resulting in a comprehensive 49-rule inference base that dynamically schedules PID gains to address aerodynamic nonlinearities. The SAC algorithm-operating as an off-policy actor-critic reinforcement learner with entropy maximization-performs online meta-tuning of the fuzzy-scaled gains, enabling continuous adaptation to turbulent wind conditions, parameter uncertainties, and unmodeled disturbances. Practical implementation challenges are mitigated through anti-windup back-calculation to prevent integral saturation and low-pass filtering to suppress derivative kick and measurement noise. Stability of the hybrid closed-loop system is analyzed using Lyapunov methods adapted for reinforcement learning frameworks, ensuring boundedness and convergence. Extensive simulations in MATLAB/Simulink on a 2 MW doubly-fed induction generator (DFIG) wind turbine under realistic IEC-compliant turbulent wind profiles demonstrate superior performance: up to 25% improvement in annual energy production (AEP), 22% reduction in fatigue equivalent loads, and markedly smoother rotor speed tracking compared to conventional PID, standalone fuzzy PID, and baseline actor-critic methods. This work extends the authors' prior SAC-based meta-tuning framework from robotic manipulator trajectory tracking [1] to renewable energy systems, offering a high-performance, learning-enabled solution for modern wind farms. achieving inference times under 5 ms on embedded platforms such as NVIDIA Jetson Nano with TensorRT optimization