ARCHIVES

This paper presents a reinforcement learning (RL)-based adaptive control optimization framework for real-time control and energy maximization of wave energy converters (WECs) under hydrodynamic and environmental uncertainties. The stochastic, nonlinear, and time-varying nature of ocean waves makes WEC control challenging, particularly when conventional controllers depend on accurate hydrodynamic models and fixed control parameters. To address these limitations, the proposed framework employs a deep actor–critic RL architecture that learns optimal control policies directly from interaction with the WEC system and dynamically adjusts power take-off (PTO) damping and control force in response to changing sea states. The control objective is formulated as a multi-objective reward function designed to maximize absorbed wave power while minimizing excessive PTO force, device displacement, and structural loading. Uncertainty-aware state representation is incorporated using real-time wave elevation, body velocity, displacement, PTO force, and estimated hydrodynamic disturbance terms. The proposed controller is evaluated using a high-fidelity point absorber WEC model under regular and irregular wave conditions, including Pierson–Moskowitz and JONSWAP spectra, with additional uncertainty introduced through ±20% variation in hydrodynamic coefficients, 5% sensor noise, and rapidly varying wave periods. Simulation results demonstrate that the proposed RL-based controller significantly improves WEC performance compared with conventional PI control and deterministic MPC. Under nominal irregular sea states, the proposed method increases average absorbed power by 28.6% compared with PI control and 14.8% compared with MPC. Under high-uncertainty sea states, the controller maintains robust operation, achieving a 24.3% improvement in energy capture, while reducing power fluctuation by 18.7% and limiting peak PTO force by 16.5% compared with MPC. In addition, the proposed approach reduces displacement constraint violations by 31.2% and improves conversion efficiency from 64.8% to 78.5% across mixed sea-state profiles. The controller also demonstrates rapid adaptation, reaching stable policy performance within 8–12 wave cycles following abrupt sea-state changes. These results confirm that the proposed RL-based adaptive optimization framework provides superior energy maximization, robustness, and real-time adaptability for WEC systems operating in uncertain marine environments. Overall, the study highlights the potential of reinforcement learning as an intelligent control solution for improving the efficiency, reliability, and commercial viability of next-generation wave energy conversion technologies.