Fatigue-Reduction Active Power Allocation in Wind Farms Using Convex Reward Reinforcement Learning

As global emphasis shifts toward renewable energy, particularly wind power, ensuring the reliability and safety of wind turbines has become imperative, especially given the significant costs associated with maintenance. This paper presents a novel approach to minimizing equivalent fatigue loads in wind farms through a Fatigue-Reduction active power allocation algorithm that leverages convex reward reinforcement learning. The proposed algorithm integrates Soft Actor-Critic (SAC) reinforcement learning with Input Convex Neural Networks (ICNN) to enhance convergence and generalization while effectively addressing the nonlinear dynamics inherent in wind farm operations. A sliding window equivalent fatigue load calculation model based on ICNN is introduced, transforming the fatigue assessment problem into a convex optimization framework, thereby improving computational efficiency. The study employs a 50MW wind farm simulation to evaluate the proposed methods, demonstrating superior performance in fatigue load reduction compared to traditional approaches. The results indicate substantial improvements in both computational speed and accuracy, underscoring the practical applicability of this methodology for optimizing wind farm operations.