Reinforcement Learning for Reliable Power Allocation and Load Mitigation in Wind Farm

Abstract

Wind energy is increasingly recognized as a key element in advancing a sustainable energy infrastructure and achieving carbon neutrality. However, the integration of wind power into the electrical grid presents significant challenges, particularly in maintaining grid frequency stability due to the variable and unpredictable nature of wind. This often necessitates precise control of power generation, which, in turn, imposes additional fatigue loads on wind turbines. Mitigating these loads is essential for lowering maintenance costs and enhancing turbine longevity. Recent advances in data-driven approaches have shown promise in optimizing power generation and grid frequency control. Notably, reinforcement learning has a natural advantage in solving complex optimization and control problems. This paper presents a novel application of RL for active power control within wind farms, aiming to devise effective power allocation strategies that minimize fatigue loads. Based on shaft torque and tower bending moment information, which are related to wind turbine fatigue, an advanced RL-based controller is designed and trained through iterative interaction with the operational environment. Furthermore, the control effectiveness is evaluated across various operational conditions. The findings confirm that the controller performs satisfactorily in mitigating fatigue loads, highlighting its viability for real-world implementation.