Optimization of Underactuated Ship Sliding Mode Controller Based on the DDPG Algorithm

Due to the development of ship intelligence, the demand for precise controllers is increasing day by day. In traditional design, the trial-and-error method often leads to the use of inaccurate controller parameters, causing oscillations when environmental disturbances are present. To address these issues, this study introduces deep learning to optimize the controller parameters, particularly those used for sliding mode controllers in underactuated ships. We adopt the Deep Deterministic Policy Gradient (DDPG) algorithm based on the Actor Critic framework. By freely setting the reward function to match the control objective, the trained model can output more accurate controller parameters, with the resulting change rates for the robustness-related parameters being 38%, 27.5%, 5%, and 200%, respectively. The results of the simulation comparison experiment show that the parameter-optimized sliding mode controller improves the path-tracking accuracy and rudder angle stability by about 20%, which is significantly better than that of the controller parameterized by the traditional trial-and-error method. It also more effectively meets the control requirements of intelligent ships.