| 引用本文： |
-
孔建国,李龙超,梁海军,等.一种基于Transformer-TCN-GRU的未来空管4D航迹预测方法[J].电讯技术，2026，66（3）： - . [点击复制]
- KONG Jianguo,LI Longchao,LIANG Haijun,et al.A Transformer-TCN-GRU-based Method for 4D Trajectory Prediction in Future Air Traffic Management[J].，2026，66（3）： - . [点击复制]
|
|
| 摘要: |
| 为满足基于航迹的运行（Trajectory-based Operations，TBO）对高精度航迹预测的严苛要求，提出了一种基于Transformer-TCN-GRU架构的4D航迹预测模型。该模型融合了Transformer的自注意力机制、时间卷积网络（Temporal Convolutional Network，TCN）的多尺度特征提取能力，以及门控循环单元（Gated Recurrent Unit，GRU）的长短期依赖建模优势，以实现对复杂航迹数据的高精度预测。在数据处理方面，通过对自动相关监视系统采集的终端区航迹数据进行插值与归一化预处理，以确保数据的连续性与稳定性。在模型训练中，采用贝叶斯优化方法对超参数进行精细调优，以进一步提升模型的预测精度与训练效率。实验结果表明，所提出的Transformer-TCN-GRU模型在4D航迹预测的精度和鲁棒性上均显著优于传统的LSTM、GRU和TCN-GRU模型，尤其在飞行状态剧烈变化的区域，其预测效果尤为突出。 |
| 关键词: 空中交通管理 4D航迹预测 基于航迹的运行（TBO） ADS-B数据 自注意力机制 |
| DOI：10.20079/j.issn.1001-893x.241111001 |
|
| 基金项目:中央高校基本科研业务费资助项目（PHD2023-035，25CAFUC10040)） |
|
| A Transformer-TCN-GRU-based Method for 4D Trajectory Prediction in Future Air Traffic Management |
| KONG Jianguo,LI Longchao,LIANG Haijun,HUANG Yujie |
| (College of Air Traffic Management，Civil Aviation Flight University of China，Guanghan 618307，China) |
| Abstract: |
| To meet the stringent requirements for high-precision trajectory prediction under trajectory-based operations(TBO)，a 4D trajectory prediction model based on a Transformer-TCN-GRU architecture is proposed.The model integrates the self-attention mechanism of Transformers，the multi-scale feature extraction capabilities of Temporal Convolutional Networks(TCN)，and the long-short term dependency modeling advantages of Gated Recurrent Units(GRU) to achieve high-precision predictions for complex trajectory data.In terms of data processing，interpolation and normalization preprocessing is applied to terminal area trajectory data collected by the Automatic Dependent Surveillance-Broadcast(ADS-B) system to ensure data continuity and stability.During model training，Bayesian optimization is employed to fine-tune hyperparameters，further enhancing both prediction accuracy and training efficiency.Experimental results demonstrate that the proposed Transformer-TCN-GRU model significantly outperforms traditional long short-term memory(LSTM)，GRU，and TCN-GRU models in terms of prediction accuracy and robustness especially in regions exhibiting drastic changes in flight states. |
| Key words: air traffic management 4D trajectory prediction trajectory-based operations(TBO) ADS-B data self-attention mechanism |
《电讯技术》编辑部 