An active-controlled heaving plate breakwater trained by an intelligent framework based on deep reinforcement learning

This paper discusses the application of Deep Reinforcement Learning (DRL) to the control of a heaving plate breakwater. It is the first time that the DRL framework is utilized to find the optimal strategy for wave dissipation. The dynamic model of the wave-plate interaction, based on an in-house Computational Fluid Dynamics (CFD) solver, is presented. After training, exciting results show that the wave dissipation performance of the active-controlled heaving plate is more outstanding than that of the passive heaving plate, especially under comparatively long period waves. Meanwhile, the availability of the control strategy to different wave conditions is proved. The Fast Fourier Transform (FFT) analyses indicate that the heaving motion of the active-controlled plate can reduce the amplitude of the fundamental component of the transmitted wave, compared to the fixed plate. Finally, the influence of the hyper-parameters on the DRL convergence rate is discussed. This study proves the potential of DRL+CFD in actively dissipating ocean waves. •This paper presents a Deep Reinforcement Learning (DRL) framework to find the optimal strategy for controlling a heaving plate to dissipate wave energies.•The agent corresponds to a decision policy controlled by an artificial neural network (ANN) that can output an action based on a given observation of the environment state.•A reward function is properly defined from the surface elevations at the lee-side of this plate, which is used to evaluate the performance of the agent.•The availability of the control strategy to different wave conditions is proved.•The agent can successfully control the heaving speed of the horizontal plate to reduce the wave transmission coefficient.