基于CNN-GRU的船舶轨迹预测  

Prediction of Ship Trajectory Based on CNN-GRU

作　　者：万洪亮 潘家财[1] 甄荣[1,2] 石自强 WAN Hong-liang;PAN Jia-cai;ZHEN Rong;SHI Zi-qiang(Navigation College,Jimei University,Xiamen Fujian 361021,China;Hubei Key Laboratory of Inland Shipping Technology,Wuhan Hubei 430063,China)

机构地区：[1]集美大学航海学院,福建厦门361021 [2]内河航运技术湖北省重点实验室,湖北武汉430063

出　　处：《广州航海学院学报》2022年第2期12-18,共7页Journal of Guangzhou Maritime University

基　　金：国家自然科学基金项目(52001134);内河航运技术湖北省重点实验室开放基金(NHHY2020001);福建省中青年教师教育科研项目(JAT190293)。

摘　　要：针对现有基于CNN、GRU及CNN-LSTM的船舶轨迹预测模型精度不高、运行时间较长等问题,提出一种基于卷积神经网络(Convolutional Neural Networks,CNN)和门控循环单元(Gated Recurrent Unit,GRU)的船舶轨迹预测混合模型(CNN-GRU).构建了基于船舶AIS信息的船舶轨迹特征表达方法,以目标船舶连续4个时刻的轨迹特征值作为输入,以第5个时刻轨迹特征值作为输出,训练构建的CNN-GRU轨迹预测网络,对未来船舶轨迹进行预测,并与现有模型进行对比.实例验证表明:CNN-GRU模型的预测精度显著提升,经度误差不超过3×10^(-5)(°),纬度误差不超过5.5×10^(-4)(°),相较于CNN-LSTM模型,预测效率显著提高,运行时间减少19.1 s.Aiming at the problems of low accuracy of the existing ship trajectory prediction models(CNN,GRU)and long running time of the hybrid prediction model(CNN-LSTM),a novel hybrid prediction model(CNN-GRU)based on convolutional neural networks(CNN)and gated recurrent unit(GRU)was proposed.In this study,the ship trajectory feature expression method is established based on ship AIS.The trajectory feature values of the target ship at four consecutive moments are used as input,and the fifth one is used as output to train the proposed CNN-GRU network to predict future ship trajectories.Compare with existing models,the experiments show that the prediction accuracy of CNN-GRU is significantly improved,the longitude error does not exceed 3×10^(-5),and the latitude error does not exceed 5.5×10^(-4).Compared with the CNN-LSTM model,the running time of the proposed model is significantly reduced by 19.1 seconds.

关 键 词：船舶轨迹预测 混合模型 卷积神经网络 门控循环单元 AIS信息 

分 类 号：U675.7[交通运输工程—船舶及航道工程]