Deep Reinforcement Learning Based Online Parameter Tuning for Active Disturbance Rejection Control and Application in Ship Course Tracking

The linear active disturbance rejection control (LADRC) has been widely used in many control fields. However, the controller with fixed parameters cannot achieve the optimal performance in an environment with compound disturbances. To achieve the online parameter tuning for LADRC, a deep reinforcement learning algorithm called the soft actor-critic (SAC), is used. Then the SAC-LADRC controller is proposed and applied in the ship course control to obtain the optimal parameters of LADRC in different states. In simulations, comparisons with the conventional LADRC controller are presented, and the effectiveness of the proposed method is verified.