基于船舶运动行为与时序图神经网络的轨迹预测研究  

Research on Trajectory Prediction Based on Ship Motion Behavior and Temporal Graph Neural Network

作　　者：魏昊坤 陈金勇[1] 刘敬一 楚博策 张文宝 姜岩松 郭琦 裴新宇 WEI Haokun;CHEN Jinyong;LIU Jingyi;CHU Boce;ZHANG Wenbao;JIANG Yansong;GUO Qi;PEI Xinyu(The 54th Research Institute of CETC,Shijiazhuang 050081,China)

机构地区：[1]中国电子科技集团公司第54研究所,石家庄050081

出　　处：《计算机测量与控制》2024年第10期39-46,共8页Computer Measurement &Control

基　　金：中国博士后科学基金项目(2021M703021);河北省重点研发计划项目(22340301D);河北省博士后基金项目(B2021003031)。

摘　　要：随着我国“海洋强国”战略的提出,航运业飞速发展;海上交通量迅猛增长,碰撞事故频发,同时积累了丰富的船舶航行数据,亟需在此数据基础上,对船舶的航行位置进行长时序的预测,加强对海域的整体交通状况的认知,降低船舶碰撞事故率;为此在实验中首先对AIS数据进行预处理,剔除其中的易于去除的异常点,提出基于船舶航行特征的动态轨迹去纠缠方法去除纠缠点;其次,依据船舶航行特征提出顾及行为语义约束的时空轨迹密度自适应聚类方法对船舶运动模式进行挖掘,得到船舶典型运动行为模式;最后,针对船舶轨迹以及船舶运动模式,提出一种基于运动模式的时序图神经网络轨迹预测模型,对轨迹进行长时序预测,选取粤港澳大湾区作为实验海域,经对比实验验证,该模型在长时序预测上效果优于传统模型。With the promotion of the“Marine Power”strategy in China,the shipping industry is rapidly developing.Rapid growth in maritime traffic leads to frequent ship collisions.There is an urgent need for the long-term prediction of ship trajectories based on accumulated ship navigation data to enhance the awareness of maritime traffic conditions and reduce collision rates.Firstly,the paper pre-processes the automatic identification system(AIS)data to eliminate easily removable outliers,and proposes a dynamic trajectory decorrelation method based on ship characteristics to remove decorrelation points.Then,a self-adaptive spatiotemporal trajectory clustering method with behavioral semantic constrains based on the characteristics of ship navigation is proposed to mine ship motion patterns and obtain typical ship behavior patterns.Finally,for the ship trajectories and motion patterns,a motion pattern-based temporal graph neural network model is presented.For long-term prediction of ship trajectories and patterns,the Guangdong-Hong Kong-Macao Greater Bay Area is selected as a test region,comparative experiments validate that the proposed model outperforms the traditional models in long-term prediction.

关 键 词：船舶自动识别系统数据 船舶轨迹预测 轨迹预处理 轨迹聚类 图神经网络 门控循环单元 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]