A PPO-based PI tuning scheme for pitch control of floating offshore wind turbines

Abstract Blade pitch control is critical for stabilizing floating offshore wind turbines (FOWTs) in above-rated regions. However, conventional proportional-integral (PI) controllers, typically derived from simplified linear models, often suffer from model mismatch and limited capability in multi-objective optimization. To overcome these limitations, this paper proposes a novel PI tuning framework driven by a proximal policy optimization (PPO) agent. Characterized by its model-free nature, the PPO agent is trained directly via real-time interaction with a high-fidelity FOWT nonlinear model under stochastic turbulent wind and irregular wave loads. Comparative simulations against benchmark gain-scheduled PI (GSPI) and individual pitch control (IPC) strategies were conducted. The results demonstrate the superior performance of the proposed strategy. It reduces generator speed root mean square error by 78% on average while maintaining robustness against turbulence variations. Furthermore, platform pitch standard deviation is reduced by 34%, with significant low-frequency attenuation. Aerodynamic thrust variation decreases by 5.3% overall without compromising the mean value, and structural loads are effectively mitigated. This study presents a robust, data-driven paradigm for FOWT control, offering a promising avenue for enhancing system resilience and efficiency in complex marine environments.