Double Deep Q Network Optimized Linear Active Disturbance Rejection Control for Ship Course Keeping

Abstract

This paper mainly proposes a parameter-optimized linear active disturbance rejection controller (LADRC) based on a double deep Q network (DDQN) and applies it to ship course control. Firstly, based on the separate mathematical models’ equation, a ship’s dynamic model is established. Then, a LADRC based course keeping controller is designed to overcome the ship’s environmental disturbances and internal uncertainty during navigation. Furthermore, to facilitate LADRC parameter adjustment and obtain a better performance of ship course keeping control, the DDQN is applied to tune the adaptive parameters of LADRC. Finally, simulation results and comparisons on ship course keeping show that the proposed DDQN optimized LADRC can control the ship’s heading angle to track the planned course, and the control performance outperforms the traditional LADRC.KeywordsShip course keeping controlLinear active disturbance rejection controlParameter optimizationDouble deep Q networkReinforcement learning