基于海上避碰规则和机器学习的辅助避碰决策方法研究  

Research on Decision-making Method of Auxiliary Collision Avoidance Based on COLREGS and Machine Learning

作　　者：冯涂超 郑茂 尹奇志[1] 初秀民[2] 谢朔 FENG Tuchao;ZHENG Mao;YIN Qizhi;CHU Xiumin;XIE Shuo(School of Energy and Power Engineering, Wuhan University of Technology, Wuhan 430080, China;Intelligent Transportation Systems Research Center, Wuhan University of Technology, Wuhan 430080, China)

机构地区：[1]武汉理工大学能源与动力工程学院,武汉430080 [2]武汉理工大学智能交通中心,武汉430080

出　　处：《武汉理工大学学报（交通科学与工程版）》2021年第1期111-116,共6页Journal of Wuhan University of Technology（Transportation Science & Engineering）

基　　金：国家重点研发计划项目(2018YFB1600400);船舶态势智能感知系统研制项目(MC-201920-X01);国家自然科学基金项目(51709220)资助。

摘　　要：为解决船舶避碰辅助决策问题,提出了一种基于国际海上避碰规则和机器学习的辅助避碰决策方法.首先建立船舶避碰场景仿真模型,收集不同态势下的船舶会遇场景样本,再依据国际海上避碰规则(COLREGs)获取规范避碰决策样本.以本船及目标船的航速和方位角为特征建立特征状态空间,分别利用BP神经网络(BP-neural network)和支持向量机(support vector machine,SVM)两种机器学习方法对所得到的辅助决策结果进行分类训练,实现在一定的相对距离内船舶辅助避碰自主决策.仿真结果显示经由BP神经网络和SVM方法建立的辅助决策模型对于不同航行态势均可提供较好的避碰决策,可用于具有自学习能力的船舶辅助避碰决策支持系统中.In order to solve the problem of aided decision-making for ship collision avoidance,an aided decision-making method for collision avoidance based on International Rules for Collision Avoidance at Sea and machine learning is proposed.Firstly,the simulation model of ship collision avoidance scene was established,and the samples of ship collision avoidance scenes in different situations were collected.Then,the standard collision avoidance decision samples were obtained according to the International Rules for Collision Avoidance at Sea(COLREGS).The characteristic state space was established based on the speed and azimuth of the ship and the target ship,and two machine learning methods,BP-neural network and Support Vector Machine(SVM),were used to classify and train the obtained auxiliary decision-making results,so as to realize the autonomous decision-making of ship auxiliary collision avoidance within a certain relative distance.The simulation results show that the aided decision-making model established by BP neural network and SVM method can provide better collision avoidance decision-making for different navigation situations,and can be used in the ship aided collision avoidance decision support system with self-learning ability.

关 键 词：机器学习 避碰决策 海上避碰规则 货运船舶 

分 类 号：U675.96[交通运输工程—船舶及航道工程]