<div style="text-align:justify"> In the weak grid environment with high penetration of new energy , the traditional PI control is not fast enough, which seriously affects the performance of the grid-connected inverter system. For this reason, this paper proposes a study of three-phase LCL-type PV grid-connected inverter control based on reinforcement learning. The original current loop is replaced with a reinforcement learning module. By adjusting the reward function in reinforcement learning and a lot of training, a agent can be obtained. Under the excitation of this agent, the grid-connected inverter system will have a better performance. Finally, the current control of PV grid-connected inverter based on reinforcement learning is verified to be better by comparing with traditional PI control in simulation.</div>

grid-connected inverter; PI control;rapidity; reinforcement learning; weak grid