基于Attenton-LSTM神经网络的船舶航行预测  

A method of ship navigation prediction based on Attenton-LSTM neural network

作　　者：徐国庆 马建文 吴晨辉 张安西 XU Guo-qing;MA Jian-wen;WU Chen-hui;ZHANG An-xi(School of Navigation,Shandong Jiaotong University,Weihai 264200,China)

机构地区：[1]山东交通学院航海学院

出　　处：《舰船科学技术》2019年第23期177-180,共4页Ship Science and Technology

基　　金：山东省交通运输科技计划资助项目(2018B70);山东交通学院研究生科技创新基金资助项目(2019YK013)

摘　　要：航行预测是无人艇关键技术之一。航行问题复杂度较高,传统的预测算法无法满足当前需求。为此,提出一种基于注意力机制-长短期记忆(Attention-Long Short Term Memory,Attention-LSTM)的多维船舶航行预测算法,结合船舶自动识别系统(Automatic Identification Systerm,AIS),采用注意力机制突出对船舶航行起关键作用的输入特征,实现对船舶未来时刻经度、纬度、航向、航速的预测。以成山角海域真实数据为例,进行仿真对比实验,结果表明所提方法具有更好的精确性和鲁棒性。One of the key technologies of unmanned warships is the prediction of ship navigation.Because the problem of ship navigation has high complexity,the traditional prediction algorithm can not meet the needs of solving this kind of problem.For this reason,a ship navigation prediction algorithm based on Attention-LSTM neural network is proposed.In this prediction algorithm,the input features which play a key role in ship navigation are highlighted by combining AIS and the attention mechanism,so as to predict the longitude,latitude,heading,speed and other factors of the ship in the future.Taking the real data of Chengshanjiao sea area as an example,the simulation experiment is carried out,and according to the results,it can be found that the proposed algorithm has better accuracy and robustness.

关 键 词：航行预测 船舶自动识别系统 误差反向传播算法 注意力机制 长短期记忆 

分 类 号：U675.9[交通运输工程—船舶及航道工程]