Deep reinforcement learning-based non-causal control for wave energy conversion

As one of the most promising renewable energy resources, ocean wave energy has not been widely commercialized compared to wind energy and solar energy due to its high Levelized Cost of Electricity (LCoE). It has been long recognized that wave energy converter (WEC) control can increase the capture width ratio and enhance the robustness of the WEC against extreme sea states. However, some rigid-body WECs have high nonlinearities and soft-body WECs such as Dielectric Elastomer Generators (DEGs)/Dielectric Fluid Generators (DFGs) can barely be precisely modeled. To tackle these challenges, this paper aims to propose an optimal control scheme that has less dependence on the dynamical model by introducing deep reinforcement learning into the foundation of a non-causal optimal control strategy. The gain parameters are adjusted adaptively in real time to account for an increasing understanding of this scheme on the WEC behavior and the incoming wave. Furthermore, by systematically contrasting outcomes obtained with various prediction time steps, this investigation aims to pinpoint the most effective prediction strategy for optimizing energy capture efficiency. The robustness of the proposed control against prediction errors and model uncertainties has been verified by using the realistic wave data gathered from the coast of Cornwall, UK. •A non-causal control scheme is proposed based on reinforcement learning.•Wave prediction is incorporated into the control scheme to increase the energy output.•A pair of real-time adjustable gain parameters have been optimized via reinforcement learning.•The proposed control algorithm is generally applicable to other WECs.•The proposed method is robust against model uncertainties and prediction errors.