基于CNN-LSTM-attention模型航迹预测研究  

Research on Track Prediction Based on CNN-LSTM-attention Model

作　　者：孔建国[1] 李亚彬 张时雨 陈超[1] 梁海军 KONG Jian-guo;LI Ya-bin;ZHANG Shi-yu;CHEN Chao;LIANG Hai-jun(Civil Aviation Flight University of China,Guanghan 618000,China)

机构地区：[1]中国民用航空飞行学院,四川广汉618000

出　　处：《航空计算技术》2023年第1期1-5,共5页Aeronautical Computing Technique

基　　金：四川省科技计划项目资助(2022YFG0210);2022年中央高校基本科研业务费项目资助(ZHMH2022-009)。

摘　　要：以航迹预测方法作为切入点,重庆-广州航路航空器记录的ADS-B数据作为研究内容,提出了一种融合注意力机制的长时序航迹预测方法(CNN-LSTM-attention)。研究运用一维卷积神经网络对航迹数据多维特征进行提取,并将经纬度、高度、速度、航向等的多维特征向量构造成时序形式作为LSTM网络输入,通过赋予LSTM网络隐含层的权重占比并区别不同时序点隐藏层信息对未来航迹预测的影响程度来达到优化预测模型的作用。构建好的CNN-LSTM-attention模型采用Adam优化算法进行训练,LSTM和CNN-LSTM作为实验对比模型,将决定系数R^(2)作为模型评价标准来衡量航迹预测模型的准确性。实验结果表明加入注意力机制的神经网络预测模型CNN+LSTM+attention(卷积神经网络-长短期记忆网络-注意力机制)的方法相较于其他两种,其预测精确性更高。This study takes the track prediction method as the entry point of this paper,ADS-B data recorded by aircraft on the Chongqing-Guangzhou route as the research content,a long time series track prediction method integrating attention mechanism(CNN-LSTM-ATTENTION) is proposed.In this paper,one-dimensional convolutional neural network is used to extract the multi-dimensional features of the track data,and the multi-dimensional feature vectors of longitude and latitude,altitude,speed and heading are constructed into the time-series form as the input of LSTM network.By assigning the weight ratio of hidden layer to LSTM network and distinguishing the influence degree of hidden layer information at different time sequence points on future track prediction,the optimal prediction model can be achieved.Adam optimization algorithm was used to train the CONSTRUCTED CNN-LSTM-Attention model.LSTM and CNN-LSTM were used as experimental comparison models.R^(2) were used as model evaluation criteria to measure the accuracy of the track prediction model.The experimental results show that the neural network prediction model CNN+LSTM+ ATTENTION(convolutional neural network-Long and short-term memory network-attention mechanism) has higher prediction accuracy than the other two methods.

关 键 词：航迹预测 CNN-LSTM-attention模型 注意力机制 ADS-B航迹数据 神经网络 

分 类 号：V355[航空宇航科学与技术—人机与环境工程] TP391[自动化与计算机技术—计算机应用技术]