基于Transformer的管制员工作负荷预测  

Transformer-based Workload Prediction for Terminal Area Controllers

作　　者：关雪琦 卢朝阳[1] 苟利鹏 张慧子 GUAN Xue-qi;LU Chao-yang;GOU Li-peng;ZHANG Hui-zi(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China)

机构地区：[1]南京航空航天大学,江苏南京211000

出　　处：《航空计算技术》2025年第1期54-58,共5页Aeronautical Computing Technique

基　　金：国家自然科学基金民航联合基金重点项目资助(U2033203)。

摘　　要：由于民航业持续高速发展,伴随着空中交通流量的快速增长,管制员的工作效率与空中交通流量之间的不平衡问题日益凸显。针对现有管制员负荷计算与预测方法存在无法准确长期预测的问题,提出一种基于Transformer模型的管制员工作负荷预测模型。该方法在循环神经网络捕捉管制员工作负荷短期依赖特征的基础上,通过编码器-解码器结构有效地捕捉了其长期依赖特征,提取并组合多个特征序列,实现对管制员工作负荷的准确预测。实验结果表明,Transformer模型相比于目前最普遍应用的SVR、随机森林、LightGBM模型,MAPE分别降低了21.47%、12.91%、9.25%,能取得更好的预测效果;该方法在管制员工作负荷预测中具有更加良好的性能和准确性。Due to the continuous rapid development of civil aviation industry,accompanied by the rapid growth of air traffic flow,the imbalance between the controller′s work efficiency and air traffic flow has become increasingly prominent.Aiming at the existing controller load calculation and prediction methods that have the problem of being unable to accurately predict in the long term,this paper proposes a new controller workload prediction model based on the Transformer model.On the basis that recurrent neural network can capture the short-term dependence of controller workload,the method captures the long-term dependence of controller workload through encoder-decoder structure,and extracts and combines multiple feature sequences to realize the accurate prediction of controller workload.The experimental results show that the Transformer model can achieve better prediction results by reducing the MAPE by 21.47%,12.91%,and 9.25%,respectively,compared with SVR,Random Forest,and LightGBM models,which are the most commonly used models at present,and the method has better performance and accuracy in the controller workload prediction.

关 键 词：空中交通管制 管制员 工作负荷预测 Transformer模型 时间序列 

分 类 号：V355[航空宇航科学与技术—人机与环境工程]