Deep Reinforcement Learning Control and Wave Tank Testing of Wave Energy Converters

The primary objective of control strategies for wave energy converters (WECs) is to optimize the conversion of wave energy into electrical power under mild wave conditions while ensuring the structural integrity and safety of the device under extreme wave conditions. Existing model-based control approaches encounter significant challenges, particularly in the accurate modeling of the system and in the prediction of wave excitation forces (WEFs). Moreover, constraints such as float displacement impose limitations on parameter selection, which in turn reduces the efficiency of energy utilization. To address these limitations, model-free control methods based on deep reinforcement learning (DRL) offer a promising alternative, as they do not require precise modeling or prediction of WEFs and can efficiently handle system constraints. In this study, we propose a DRL-based control method for direct-drive point absorber WECs, utilizing the soft actor-critic (SAC) algorithm. The trained control strategies were applied to the WEC and validated through wave tank experiments. The results demonstrate that the DRL-based control strategy significantly enhances energy capture in both regular and irregular wave conditions, while strictly adhering to displacement constraints. These findings highlight the potential of the control strategy proposed as an effective tool for improving both the efficiency and safety of WEC systems.