Fuzzy-based collective pitch control for wind turbine via deep reinforcement learning

Area of Science:

• Renewable Energy Systems
• Control Engineering
• Artificial Intelligence in Engineering

Background:

• Wind turbines (WTs) exhibit complex nonlinear dynamics and uncertainties, particularly in region 3, challenging effective control design.
• Aerodynamic complexity, mechanical factors, and wind fluctuations necessitate robust control strategies for optimal performance.
• Existing control methods often struggle to adapt to the inherent uncertainties and nonlinearities of WTs.

Purpose of the Study:

• To develop a novel model-free reinforcement learning (RL) collective pitch angle controller for efficient operation of WTs in region 3.
• To enhance generator speed stability, maximize power output, and minimize fluctuations under varying wind conditions and system uncertainties.
• To ensure controller robustness and generalization across diverse operating conditions and WT dynamics.

Main Methods:

• A deep deterministic policy gradient (DDPG) algorithm was employed to train multiple RL agents in a medium-fidelity WT environment.
• Imitation learning was utilized for initial efficient sample collection to accelerate training convergence.
• A fuzzy system integrated the outputs of multi-trained DDPG agents to create a smooth, adaptive fuzzy DDPG (F-DDPG) controller.

Main Results:

• The proposed F-DDPG controller demonstrated superior performance in stabilizing generator speed and maximizing power output compared to GSPI, LQR, and single-DDPG controllers.
• Simulations in high-fidelity onshore and offshore 5-MW WT environments (OpenFAST/MATLAB) confirmed the controller's robustness and generalization capabilities.
• The F-DDPG controller effectively managed system uncertainties, nonlinearities, and pitch limits, ensuring optimal operation across different mean wind speeds.

Conclusions:

• The F-DDPG controller offers a robust and adaptive solution for wind turbine pitch control in region 3, outperforming traditional and single-agent RL methods.
• The model-free RL approach combined with fuzzy logic provides significant advantages in handling the complex dynamics and uncertainties of wind turbines.
• This research paves the way for more efficient and reliable wind energy generation through advanced AI-driven control strategies.