面向船舶智能航行的多目标轨迹预测算法  

Multi-target trajectory prediction algorithm for ship intelligent navigation

作　　者：徐海祥[1,2] 卢烨彬 冯辉[1,2] 周志杰 XU Haixiang;LU Yebin;FENG Hui;ZHOU Zhijie(Key Laboratory of High Performance Ship Technology of Ministry of Education,Wuhan University of Technology,Wuhan 430063,China;School of Naval Architecture,Ocean and Energy Power Engineering,Wuhan University of Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学高性能船舶技术教育部重点实验室,湖北武汉430063 [2]武汉理工大学船海与能源动力工程学院,湖北武汉430063

出　　处：《华中科技大学学报（自然科学版）》2024年第10期54-59,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(51979210,52371374)。

摘　　要：针对目前预测算法中船舶轨迹样本分布不均衡、群体交互关系利用率低及预测结果不符合船舶运动学特性等问题,提出了一种基于稀疏图卷积网络(S-GCN)的多目标轨迹预测算法.首先,设计了可学习的非概率采样网络(NPSN)以生成分布均衡的轨迹样本;其次,基于船舶个体与群体的关系提出了船舶集群表示方法以推断符合海事规则的多目标交互方式;最后,采用交互式多模型(IMM)状态估计算法对预测轨迹进行了滤波修正以满足船舶运动学机理.实验结果表明:所提出的算法性能有了较大提升,平均位移误差(ADE)和终点位移误差(FDE)分别为17.06 m和29.49 m,优于S-GCN和其他现有的常用预测算法.Regarding the current issues of imbalanced sample distribution,low utilization of group interaction relationships,and prediction results not conforming to vessel kinematics in trajectory prediction algorithms,a multi-objective trajectory prediction algorithm based on Sparse graph convolutional networks(S-GCN) was proposed.First,a learnable non-probabilistic sampling network(NPSN) was designed to generate trajectory samples with balanced distributions.Then,a method for representing vessel clusters was proposed based on the relationships between individual vessels and groups to infer multi-objective interaction modes that comply with maritime rules.Finally,an interactive multiple model(IMM) state estimation algorithm was employed to filter and correct predicted trajectories so as to satisfy vessel kinematic principles.Experimental results show significant improvements in algorithm performance,with average displacement errors(ADE) and final displacement errors(FDE) of 17.06 m and 29.49 m,respectively,outperforming S-GCN and other commonly used prediction algorithms.

关 键 词：智能船舶 多目标轨迹预测 稀疏图卷积网络 非概率采样网络 集群表示 滤波修正 

分 类 号：U675.7[交通运输工程—船舶及航道工程] TN953[交通运输工程—船舶与海洋工程]