Prediction-Error-Compensated Model Predictive Control with a Forgetting Factor for Ship Trajectory Tracking and Collision Avoidance

Abstract

To improve the prediction reliability of model predictive control under model mismatch and environmental disturbances, this paper proposes a prediction-error-compensated MPC method with a forgetting factor for ship trajectory tracking and collision avoidance. The proposed method is developed within a Frenet-frame virtual ship group framework. A compensation term is introduced to correct the predicted heading-output sequence in the MPC prediction horizon, and a forgetting factor is used to regulate the influence of historical prediction errors on the current compensation term. Candidate trajectories are generated in the Frenet frame and evaluated using a hierarchical cost function to select a feasible obstacle avoidance path. Simulation experiments were conducted using the MMG model of the “Yukun” ship in three types of maritime encounter scenarios (head-on encounter, crossing encounter, and overtaking encounter) and a comprehensive obstacle scenario. The simulation results show that the proposed method can generate feasible local collision avoidance trajectories in complex obstacle environments and typical encounter scenarios, and maintain good trajectory tracking performance. It provides an effective scheme for trajectory tracking and collision avoidance of underactuated ships with great theoretical and practical value.