Abstract: In order to tackle of the issue for the poor navigation stability of unmanned surface vehicle (USV) under interference conditions, an intelligent adjustment strategy for control parameters based on the deep reinforcement learning method is proposed to control of the USV effectively under interference conditions. [Methods] The dynamical model of the USV, which combines the line-of-sight (LOS) method and the PID navigation controller, is established in order to conduct its navigation control task. In view of the time-varying characteristics of the PID parameters for the course control under the conditions of interference, the deep reinforcement learning theory is introduced. The environmental state, action and reward function of the intelligent agent are designed to adjust the PID parameters online. An improved DDPG algorithm is proposed to increase the convergence speed as well as to tackle of the issue for the occurrence of local optima during the training process. Specifically, the original experience pool is separated into successful and failure experience pools and an adaptive sampling mechanism is designed in order to optimize the experience pool playback structure. [Results] Simulation experiments show that the improved algorithm converges rapidly, and the average return is slightly improved in the later stage of training. Under the condition of external disturbance, the lateral errors and the heading angle deviations of the controller based on the improved DDPG algorithm are reduced significantly. Path tracking can be maintained more steadily after fitting the desired path faster. [Conclusions] The improved algorithm greatly reduces the cost of training time, enhances the steady-state performance of the agent in the later stage of training, and achieves more accurate tracking of the path.