Optimization of Offshore Wave Energy Harvester Using Reinforcement Learning

Wave energy has been of interest for its sustainability and high energy storing capability. However, the complexity of ocean waves caused by many environmental factors makes the harnessing process without sophisticated surface measurements difficult and inefficient.  In particular, most existing wave energy converter systems rely on static power take-off technologies, but active control of the power take-off (PTO) can enhance energy extraction. This project aims to utilize reinforcement learning to actively control a PTO damping coefficient to optimize the power over a specified time horizon with only one causal surface wave measurement.  It is achieved by integrating the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm with the PTO system of a two-body point absorber device. The algorithm learns the optimal damping coefficient based on the history of the time dynamics and limited knowledge of the future state based on the current wave detection sensor capabilities. The performance of the proposed control technology is tested for several wave energy concepts made up of multiple point absorbers.