摘要
针对当前飞行轨迹预测与异常检测大多依赖历史轨迹数据,未充分考虑气象条件及空域管制等外部因素,导致预测精度不足和异常检测能力的问题,本文提出了一种结合卷积神经网络(Convolutional Neural Network,CNN)、双向长短期记忆网络(Bidirectional Long Short-Term Memory,Bi-LSTM)和Transformer的飞行轨迹预测方法,即CBT-Net(CNN-BiLSTM-Transformer Network)。利用ADS-B轨迹数据,融合气象信息及空域管制数据,构建多源数据驱动的轨迹预测模型,并通过3σ统计误差分析与自监督学习结合图神经网络进行异常检测优化。实验结果表明,Bi-LSTM在短期轨迹预测中表现最佳,Transformer在长时间跨度的预测任务中优于传统线性回归模型。结合图神经网络(Graph Neural Network,GNN)的自监督学习方法在异常检测中识别能力较高,能够有效检测轨迹偏离、异常爬升/下降、速度突变等异常行为,为智能空管、无人机交通管理和航空安全监控提供技术支持。
In view of the fact that the current flight trajectory prediction and anomaly detection methods mostly rely on historical trajectory data,and do not fully consider external factors such as meteorological conditions and airspace control,which may lead to insufficient prediction accuracy and abnormal detection capabilities,this paper proposes a flight trajectory prediction method combining Convolutional Neural Network(CNN),Bidirectional Long Short-Term Memory(Bi-LSTM)and Transformer,namely CBT-Net(CNN-BiLSTM-Transformer Network).Using ADS-B trajectory data,fused meteorological information and airspace control data,a multi-source data-driven trajectory prediction model is constructed,and abnormal detection and optimization are performed through 3σstatistical error analysis and self-supervised learning combined with graph neural network.Experimental results show that Bi-LSTM performs best in short-term trajectory prediction,and Transformer is better than traditional linear regression models in long-span prediction tasks.Combined with the self-supervised learning method of Graph Neural Network(GNN),the self-supervised learning method has high recognition capabilities in abnormal detection,and can effectively detect abnormal behaviors such as trajectory deviation,abnormal climb/descent,speed mutation,and provide technical support for intelligent air traffic management,drone traffic management and aviation safety monitoring.
作者
徐英康
王杰
李振星
周丽平
张愧松
孙仁诚
XU Yingkang;WANG Jie;LI Zhenxing;ZHOU Liping;ZHANG Kuisong;SUN Rencheng(College of Computer Science&Technology,Qingdao University,Qingdao 266071,China)
出处
《青岛大学学报(工程技术版)》
2025年第1期32-40,63,共10页
Journal of Qingdao University(Engineering & Technology Edition)
关键词
飞行轨迹预测
异常检测
深度学习
人工智能
时空数据融合
flight trajectory prediction
anomaly detection
deep learning
artificial intelligence
spatiotemporal data fusion
作者简介
第一作者:徐英康(1999-),男,硕士研究生,主要研究方向为大数据分析;通信作者:孙仁诚(1977-),男,教授,主要研究方向为基于复杂网络的大数据分析。Email:src@qdu.edu.cn。
