Energy-efficient control of a thruster-assisted position mooring system using neural Q-learning techniques

Abstract

ABSTRACT Thruster-assisted position mooring (PM) systems use both mooring lines and thrusters to maintain the position and heading of marine structures in ocean environments. In order to operate in an energy-efficient manner in moderate sea conditions, appropriate setpoints need to be found for the feedback controller, where the mooring system counteracts the main environmental loads and the thrusters reduce the oscillatory motion of the marine structure. The theory of reinforcement learning (RL) provides powerful and effective tools for designing decision making agents, which can optimise their behaviours based on the interactions with the environment. We propose several designs of decision making agent based on state-of-art neural Q-learning techniques. The simulation results show that the RL agents can successfully identify the optimal setpoints through the interaction with an unknown and stochastic environment, and double Q-learning and prioritised replay techniques prove to be fairly effective in shortening the learning time of the agent.