基于 PN 和 CNN-LSTM-ATT 的航班延误预测-《计算机技术与发展》

文章信息/Info

作者:: 吴涔¹; 叶宁¹; 2 ; 王甦¹ ; 季翔宇¹; 1. 南京邮电大学计算机学院、软件学院、网络空间安全学院,江苏南京 210023;
2. 江苏省无线传感网高技术研究重点实验室,江苏南京 210023

Author(s):: WU Cen1 ; YE Ning1; 2 ; WANG Su1 ; JI Xiang-yu1; 1. School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;
2. Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks,Nanjing 210023,China

Keywords:: flight ground service; service process analysis; delay prediction; Petri Net; CNN-LSTM-ATT

摘要:: 航班延误预测对提高机场地面保障效率具有重要参考意义。针对目前航班地面保障流程复杂多变以及航班过站延误预测精度不高的问题,提出了一种基于 Petri Net 和融合预测模型 CNN-LSTM-ATT 的航班延误预测模型。首先,根据机场航班实际地面保障流程抽象构建离港航班地面保障作业 Petri Net 模型,获取保障流程中的关键作业时长成为动态特征;其次,将动态特征、航班信息、延误信息和天气信息输入 CNN-LSTM-ATT 模型中进行特征提取和分类预测,模型中引入注意力机制,通过注意力权重突出关键数据信息的影响,进一步挖掘重要特征之间的内部规律。实验结果显示,该融合模型准确率相比独立模型提升了 6% ,达到 98. 1% 。通过对不同模型的对比表明该模型能较好地应对场面流程变化并且具备较好的延误预测能力

Abstract:: Flight delay prediction is an important reference to improve the efficiency of airport ground service. Nowadays, the flightground service process is complex and changeable, and the prediction accuracy of flight transit delay?
is not high, thus a flight delayprediction model based on Petri net and fusion prediction model CNN - LSTM-ATT is designed. Firstly, the Petri Net model of thedeparting flights ground service process was abstractly constructed according to the actual ground service process of flights,and the criticalservice operation duration became the dynamic feature. Secondly,the dynamic feature,flight information,delay information and weatherinformation are input into the CNN-LSTM-ATT model for feature extraction and classification prediction. The attention mechanism isintroduced to highlight the influence of key data through attention weight and further mine the internal laws of important features. Theresults show that the fusion model improves the accuracy by 6% compared with the independent model,and the final accuracy of the experiment reaches 98. 1% . The comparison of different models shows that the model can better cope with the changes of scene and hasbetter delay prediction ability.