一种基于Attention-TCN-GRU的船舶轨迹预测模型  

A Ship Trajectory Prediction Model Based on Attention-TCN-GRU

作　　者：郑元洲[1,2] 黄海超 钱龙 曹婧欣 侯文波 李鑫 ZHENG Yuanzhou;HUANG Haichao;QIAN Long;CAO Jingxin;HOU Wenbo;LI Xin(School of Navigation Wuhan University of Technology,Wuhan 430063,China;Hubei Key Laboralory of Inland Shipping Technology,Wuhan 430063,China)

机构地区：[1]武汉理工大学航运学院,武汉430063 [2]内河航运技术湖北省重点实验室,武汉430063

出　　处：《武汉理工大学学报(交通科学与工程版)》2025年第2期439-447,共9页Journal of Wuhan University of Technology(Transportation Science & Engineering)

基　　金：国家自然科学基金(52171350)。

摘　　要：本文提出了一种串行Attention-TCN-GRU的轨迹预测模型.通过数据清洗和异常值处理等过程筛选出有效AIS数据,并采用三次样条插值算法补全船舶轨迹缺失值,有效提高数据的可用性.该模型将时间卷积神经网络(TCN)强大的时序数据特征提取能力与门控循环网络(GRU)相结合,通过串行结构设计,有效提高了船舶航行信息的处理能力.同时针对内河船舶在桥区水域及大角度弯曲航道的航行特点,将注意力机制引入预测模型,实现了较高精确度的航迹数据特征提取和趋势预测.本文开展了基于AIS数据的多工况轨迹预测实验,结果表明:Attention-TCN-GRU对内河复杂水域船舶航迹预测精确度明显优于传统神经网络.A serial Attention-TCN-GRU trajectory prediction model was proposed.Effective AIS data were screened out through data cleaning and outlier processing,and the missing values of ship trajec-tory were supplemented by cubic spline interpolation algorithm,which effectively improved the availa-bility of data.This model combined the powerful time series data feature extraction ability of time convolution neural network(TCN)with the gated cyclic network(GRU),and effectively improved the processing ability of ship navigation information through serial structure design.Meanwhile,ac-cording to the navigation characteristics of inland river ships in the waters of the bridge area and the large-angle curved channel,the attention mechanism was introduced into the prediction model,and the track data feature extraction and trend prediction with high accuracy were realized.In this paper,a multi-condition trajectory prediction experiment based on AIS data was carried out.The results show that the accuracy of Attention-TCN-GRU in predicting the ship's trajectory in complex inland waters is obviously better than that of traditional neural network.

关 键 词：船舶轨迹预测 AIS数据 时间卷积神经网络 注意力机制 Attention-TCN-GRU 

分 类 号：U698[交通运输工程—港口、海岸及近海工程]