基于注意力机制与CNN⁃GRU模型的飞机航迹预测研究  

Trajectory prediction of aircraft based on attention mechanism and CNN⁃GRU model

作　　者：徐富元[1] 蒋明 王志印 秦晋 郑子扬[1] Xu Fuyuan;Jiang Ming;Wang Zhiyin;Qin Jin;Zheng Ziyang(No.8511 Research Institute of CASIC,Nanjing 210007,Jiangsu,China;State Grid Hengyang Power Supply Company,Hengyang 421001,Hunan,China)

机构地区：[1]中国航天科工集团8511研究所,江苏南京210007 [2]国网衡阳市供电公司,湖南衡阳421001

出　　处：《航天电子对抗》2022年第6期25-30,59,共7页Aerospace Electronic Warfare

摘　　要：随着当前空中飞机数量的日益增加,有限的空域资源随之日趋紧张,针对现有航迹预测模型无法充分挖掘航迹数据中的关键价值信息的问题,提出了一种结合注意力机制与卷积神经网络(CNN)和门控循环单元(GRU)的飞机航迹预测模型。该模型以历史飞机航迹的经度、纬度、高度数据为基础,首先利用CNN模块从航迹数据中进行空间特征提取,然后将特征输入到GRU模块中进行时间特征的提取,最后利用注意力模块为特征的不同部分赋权,进一步提高模型的预测性能。实验结果表明,相较于CNN、GRU、CNN⁃GRU等模型,结合注意力机制的CNN⁃GRU模型在评价指标上有更好表现,能更精确地进行航迹预测。With the increasing number of aircrafts in the air,the limited airspace resources are becoming in⁃creasingly tense.In view of the problem that the existing track prediction model can not fully mine the key value information,an aircraft track prediction model combining attention mechanism,convolutional neural network(CNN)and gate recurrent unit(GRU)is proposed.The model used the longitude,latitude and altitude data of the historical aircraft track.Firstly,the CNN module is used to extract the spatial features from the track data,and then the features are input into the GRU module to extract the temporal features.Finally,the attention mod⁃ule is used to weight different parts of the features to further improve the prediction performance of the model.The experimental results show that compared with CNN,GRU,CNN-GRU models,the CNN-GRU model combined with attention mechanism has better performance in the evaluation index and can predict the track more accurately.

关 键 词：轨迹预测 注意力机制 卷积神经网络 门控循环单元 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]