Abstract: A large proportion of new energy and random loads are connected to microgrids, and the large disturbance caused by its uncertainty has a negative impact on the stability of DC bus voltage. In order to realize fast adaptive voltage regulation under large disturbance, a new compound control strategy based on Kalman filter (KF) and deep reinforcement learning is proposed for dual-active bridge (DAB)DC-DC converter. A deep deterministic strategy gradient reinforcement learning agent based on the Actor-Critic architecture is designed. The best observation results of KF are used as feed-forward compensation to improve the accuracy of output voltage regulation. The control parameters of DAB converter are automatically adjusted through online learning to ensure that the DC converter is stable in the face of various system disturbances. Finally, the effectiveness of the control strategy is verified by simulation and experiments.