Here, least-squares policy iteration, a reinforcement learning algorithm, is applied to the reactive control of a wave energy converter for the first time. Simulations of a linear point absorber are used for this analysis. The focus of this study is on the implementation of displacement constraints. The use of a penalty term is effective in teaching the controller to avoid the selection of combinations of the damping and stiffness coefficients that would result in excessive displacements in particular sea states. However, the controller can learn that the actions are bad only after trying them, as shown by the simulations. For this reason, a lower-level control scheme is proposed, which changes the sign of the controller force based on the magnitude of the float displacement and sign of its velocity. Its effectiveness is proven in both regular and irregular waves, although greater care is required for the determination of soft constraints.
展开▼
