基于强化学习的锚泊辅助动力定位系统智能决策研究  

Intelligent setpoint control of thruster-assisted postion mooring of a semi-submersible platform based on reinforcement learning

作　　者：余尚禹 王磊[1,2] 李博 衣凡[1] YU Shangyu;WANG Lei;LI Bo;YI Fan(State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Collaborative Innovation Center for Advanced Ship and Deep-Sea Exploration,Shanghai 200240,China)

机构地区：[1]上海交通大学海洋工程国家重点实验室,上海200240 [2]高新船舶和深海开发协同创新中心,上海200240

出　　处：《海洋工程》2019年第6期49-61,共13页The Ocean Engineering

基　　金：工信部课题“第七代半潜平台模型试验及技术研究”系泊定位技术([2016]546);国家重点研发计划(2016YFC0303405)

摘　　要：针对半潜平台锚泊辅助动力定位系统的最优定位点问题,设计了基于强化学习中深度神经网络的Q学习(DQN)控制策略的锚泊辅助动力定位的智能决策系统。该决策系统中DQN方法与比例—积分—微分(PID)控制方法相结合使用,实现系统优化。在基于机器人操作系统(ROS)平台的动力定位时域模拟程序中进行数值仿真,仿真结果验证了该系统在定位点决策问题上的可靠性和有效性,从而使半潜平台在面对未知海况时,均能寻找到最优定位点,在保证锚泊辅助动力定位系统可靠性的同时降低功率消耗,提高经济性。To improve the working performance of the thruster-assisted position mooring(PM)system,an intelligent setpoint decision strategy based on reinforcement learning is proposed.The Deep Q network(DQN)is used in combination with the PID control to achieve the optimization of the PM system.The simulation results based on ROS platform show that when operating in moderate sea conditions,the proposed intelligent setpoint decision strategy can successfully identify the optimal setpoint at which the power consumption of the thrusters can keep at a relatively low level.

关 键 词：动力定位系统 强化学习 人工神经网络 设点定位 

分 类 号：P751[交通运输工程—港口、海岸及近海工程]