基于强化学习的锚泊辅助动力定位系统定位点选取方法  

A setpoint selection method for a thruster-assisted position mooring system based on reinforcement learning

作　　者：蒋旭[1] 王磊[1] 王一听 JIANG Xu;WANG Lei;WANG Yiting(State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学海洋工程全国重点实验室,上海200240

出　　处：《海洋工程》2024年第6期41-51,共11页The Ocean Engineering

基　　金：第七代超深水钻井平台(船)创新专项。

摘　　要：海上作业中,适当的定位点可以显著降低锚泊辅助动力定位(PM)系统的能耗。为了找到能效最优的定位点,使得锚泊系统能够补偿主要的环境载荷而推进器只需抑制海洋结构物的摇荡运动,提出了一种基于模型的强化学习(RL)方法进行最优定位点的决策。该方法通过直接和间接学习更新Q函数,并且通过支持向量回归来近似环境模型的奖励函数。仿真结果表明,该方法能够在未知和随机环境中通过持续的规划、执行和学习成功地确定最优定位点,且可以有效加快决策代理的学习速度。Appropriate setpoints for the thruster-assisted position mooring(PM)systems can significantly reduce energy consumption of the thrusters in offshore operations.To identify the most energy-efficient mooring points,which allow the mooring system to compensate for the major environmental loads while the thrusters only need to mitigate the oscillatory motion of marine structures,a model-based reinforcement learning approach for the optimal positioning decision-making is proposed.This method updates the Qfunction through both direct and indirect learning,and approximates the reward function of the environmental model using support vector regression.Simulation results indicate that this approach can successfully determine the optimal setpoints in unknown and random environments through continuous planning,execution,and learning,and can effectively accelerate the learning pace of the decision-making agent.

关 键 词：锚泊辅助动力定位系统 定位点优化 强化学习 基于模型 导航控制 

分 类 号：P751[交通运输工程—港口、海岸及近海工程]