Optimal setpoint learning of a thruster-assisted position mooring system using a deep deterministic policy gradient approach

摘要

Thruster-assisted position mooring (PM) systems use both mooring lines and thrusters for station keeping of marine structures in ocean environments. To operate in an energy-efficient manner in moderate sea conditions, setpoints need to be appropriately chosen for the setpoint controller, so that the mooring system counteracts main environmental loads, while the thrusters reduce oscillatory motions of the marine structure. In this paper, reinforcement learning is used to design a decision-making agent for setpoint selection. In particular, a deep deterministic policy gradient (DDPG) approach is adopted with the powerful actor-critic architecture to continuously modify the setpoint setting at an optimal position. Extensive numerical experiments demonstrated that with the DDPG-based PM system, the intelligent agent is able to successfully identify the optimal positioning region in an unknown and stochastic environment, and the power consumption of the thrusters is maintained at a considerably low level.