基于Bi-BPNNs的船舶轨迹修复模型-《集美大学学报（自然科学版）》

文章信息/Info

Author(s):: LI Yue1; YUAN Zhi1; 2; LIU Yi2; (1.College of Technology,Hubei Engineering University,Xiaogan 432000,China;2.School of Navigation,Wuhan University of Technology,Wuhan 430063,China)

关键词:: 为了解决船舶轨迹数据的异常和丢失问题; 辅助轨迹复原和情景推演; 提出一种双向学习模型; 用于修复船舶轨迹。从AIS（Automatic Identification System）报文中提取航行船舶的上下文轨迹特征向量; 改进基础BP（Back Propagation）神经网络的拓扑结构; 构建具有双向预测功能的BP神经网络（Bi-BPNNs）模型; 修复缺失的轨迹数据。使用长江干线航行船舶的真实轨迹数据对构建的模型进行验证和分析; 通过与基础BP网络和常用线性插值方法的对比; 证明BiBPNNs模型在船舶轨迹修复上具有更好的性能和效果。

Keywords:: ship traffic flow; trajectory repair; Bi-BPNNs; bi-directional prediction

Abstract:: In order to solve the issues of trajectory data abnormality and losses,as well as to assist trajectory restoration and scenario deducing,a two-way learning method for ship trajectory data repair is proposed.Which can solve the problem of trajectory data abnormality and loss,and assist trajectory restoration and scenario inference,so as to ensure ship traffic flow analysis and marine accident investigation.Specifically,the context trajectory feature vectors of ships are extracted from AIS (automatic identification system) data,the topological structure of BP (Back Propagation) neural network is improved,and the Bi-directional BP neural network (Bi-BPNNs) model is constructed to repair the missing trajectory data.Finally,the real trajectory data of the ships sailing on the Yangtze River are collected for verification of the proposed model.Whats more,compared with basic BP network and common linear interpolation methods,the Bi-BPNNs model proposed in this paper has better performance and effect on ship trajectory data repair.