Reinforcement Learning Control of Direct-Driven Wave Energy Converter Incorporating Constrained Field Weakening Optimization

Wave energy converters have recently attracted significant attention in control methodology. The complex and variable marine environment makes modeling difficult, thus requiring model-free control methods. To maximize power capture while satisfying physical constraints, this study proposes a model-free control method based on Deep Q-Network (DQN), incorporating field weakening optimization (FWO) mechanism. The DQN-based control primarily optimizes the damping and stiffness coefficients, then the field weakening optimization determines the optimal output under linear/nonlinear constraints. Experimental results show that the proposed FWO enhances power capture by 37%-65% under various sea conditions while extending the operational speed range by 10%. Moreover, compared with resistive control and MPC, the DQN-FWO method improves power capture by 330% and 70% respectively, demonstrating validated effectiveness for maximal power extraction.