Journal Article

Reinforcement Learning-Based Individual Blade Pitch Control for Wind Turbine Fatigue Mitigation

Capaldo, Matteo, Carton, Florence, Lataillade, Tristan de

2026

Overview

In this work, we explore the use of Reinforcement Learning (RL) to learn Individual Pitch Control (IPC) functions directly from simulation, using only high-level turbine state as input. Our goal is to learn a parametric pitch signal that minimizes structural load variation based on the current operating condition. We treat this as a single-step control problem, trained using a physical simulator. We show that this approach learns effective IPC functions with high sample efficiency and produces interpretable, periodic control signals of arbitrary complexity that generalize across operating points. Results under uniform and turbulent inflow with RL-IPC control show a significant reduction in fatigue loads compared to baseline Collective Pitch Control (CPC) and it also performs better than classical Proportional Integral PI-IPC control. Our results suggest that reinforcement learning can serve as a practical tool for simulation-driven control design in physical systems with a solution that is unbiased by human intervention, relying solely on a raw performance evaluation.

• Publisher: IOP Publishing
• Subject: Load fluctuation / Performance evaluation / Pitch (inclination) / Proportional integral / Wind turbines