As an important component of modern power systems, DC microgrids are susceptible to disturbances at the load-side interface converters due to the randomness and uncertainty of renewable energy sources, resulting in poor output characteristics. In order to effectively mitigate the adverse effects of uncertainty on system performance when the controller parameters are kept constant, this paper proposes an active disturbance rejection control method based on the proximal policy optimization (PPO) algorithm. In this method, a PPO agent interacts with the traditional active disturbance rejection control system environment to perceive changes in system states and optimizes the control strategy based on feedback from a reward. During the training process, the agent explores various control actions to adaptively tune observer parameters, thereby ensuring the stability of the converter output voltage. Finally, the proposed PPO-LADRC is compared through digital simulations with the traditional linear active disturbance rejection control (LADRC) and double-closed-loop proportional-integral control methods. The results verify that the proposed control strategy can significantly improve the dynamic performance of the system under various disturbances.