Classification prediction of air traffic flow based on CNN-GRU model driven by multi-source data

Abstract

To improve the prediction accuracy of traffic flow in terminal area, a traffic flow prediction model based on dif-ferent weather scenarios (TFPM-DWS) is proposed in this paper. First, a feature set for predicting traffic flow is extracted from terminal data of traffic demand, weather, and flow control strategy. Then, convolutional neural network (CNN) and K-Means++ algorithm are used to get the embedded spatial features and cluster weather avoidance field (WAF) images into some scenarios. Based on different scenarios, TFPM-DWS is constructed by using CNN and Gated Recurrent Unit (GRU). Finally, the proposed model is validated on the historic traffic data of Baiyun Airport. Its weather data is clustered into mild weather scenarios and severe weather scenarios, and the traffic flow characteristics under the two scenarios are analyzed. Further, traffic flow predictions at 1-hour intervals are performed for the traffic flows under different weather scenarios. The comparative experimental results show that the proposed traffic flow prediction model based on different weather scenarios has higher prediction accuracy than other existing traffic flow prediction methods for terminal areas.