Abstract: Abstract: To address the critical challenges of severe data missingness, high uncertainty in influencing factors, and limited prediction accuracy in flight ground service milestone event prediction, a prediction framework named CTrans-ATGCN was proposed, which included two core components: data imputation (Convolutional Transformer, CTrans) and milestone event prediction (Attention-based Temporal Graph Convolutional Network, ATGCN). First, based on the Transformer architecture, a data imputation method, CTrans, was constructed by integrating a convolutional self-attention mechanism. A convolutional layer was introduced to extract the temporal and topological features inherent in flight ground service milestone event data, and a self-attention mechanism was employed to model potential global dependencies among different milestone events. Building on this, a milestone event prediction method, ATGCN, was constructed. The Temporal Graph Convolutional Network (T-GCN) was utilized to capture spatio-temporal dependencies in historical milestone event data. By combining an attention mechanism with a scoring function, the weights between different milestone event nodes were dynamically adjusted to achieve high-precision predictions for ground service milestone events. Finally, experimental results on a 2023 real-world dataset from an airport in North China showed that in a non-random missing scenario with a 70% missing rate, the Mean Absolute Error (MAE) of the imputation method was reduced by approximately 11.70% compared to the suboptimal GraphSAGE model. Furthermore, the “imputation + prediction” approach yielded a prediction performance comparable to the state-of-the-art SBDGCNM method on low missing rate datasets. On datasets with a 70% missing rate, MAE and RMSE values of 14.76 and 18.42 were achieved, respectively, outperforming all benchmark methods. An effective method was provided for addressing the flight ground support prediction problem under data missingness, and reliable support was offered for the efficient scheduling of ground staff and equipment.

Key words: Computer applied technology, Flight ground support, CTrans-ATGCN, Time series imputation, Milestone event prediction, Transformer model, Attention mechanism

摘要： 摘 要: 针对航班地面保障里程碑事件预测面临的数据缺失严重、影响因素不确定性强以及预测精度不高等问题，论文提出了一种航班地面保障里程碑事件预测框架：CTrans-ATGCN，包括缺失数据填补（Convolutional Transformer ，CTrans）和里程碑事件预测（Attention-based Temporal Graph Convolutional Network，ATGCN）两大核心组件。首先，基于Transformer架构，引入卷积层提取航班地面保障里程碑事件数据中蕴含的时序和拓扑特征，采用自注意力机制建模不同里程碑事件之间潜在的全局依赖关系，构建了融合卷积自注意力机制的航班地面保障数据填补方法：CTrans。在此基础上，构建了里程碑事件预测方法：ATGCN。该方法利用时序图卷积网络（T-GCN）来捕捉里程碑事件历史数据中蕴含的时空依赖关系，并通过结合注意力机制与评分函数，动态地调整不同里程碑事件节点之间的权重，从而实现对地面保障里程碑事件的高精度预测。最后，在华北地区某机场2023年真实数据集上的实验结果表明，填补方法在缺失率为70%的非随机缺失场景下，较次优模型GraphSAGE的MAE降低了约11.70%；“填补+预测”方法在低缺失率数据集上取得了与领域先进方法SBDGCNM相当的预测效果，在缺失率为70%数据集下的MAE和RMSE分别为14.76和18.42，优于所有基准方法。该研究为解决数据缺失下的航班地面保障预测问题提供了有效方法，为地勤高效调度人员与设备提供了可靠支持。

关键词: 计算机应用技术, 航班地面保障, CTrans-ATGCN, 时间序列填补, 里程碑事件预测, Transformer模型, 注意力机制