Airport surface traffic situation prediction method based on improved CNN-LSTM

In order to reduce the occurrence of aircraft waiting and conflicts, a prediction method combining time series features and convolutional neural network/long short-term memory (CNN-LSTM) network attention mechanism is proposed. Design. Appropriate spatiotemporal factors influencing traffic patterns were first selected based on ground traffic data, and then a CNN-LSTM network model was developed. The experimental outcomes indicate that the proposed model had an average absolute error of 0.7032, 0.7387, 1.0102 and 3.6534 for airport departure flow, departure queue length, airport surface traffic (AST) density and departure flight taxiing time, respectively. The root mean square errors were 1.0658, 0.9562, 1.2437, and 4.9242, respectively. The predicted values of traffic situation were between [23.51, 1.57], with significant fluctuations. Therefore, the proposed model was able to accurately predict the AST situation, improve flight operation efficiency and contribute to the healthy growth of the civil aviation industry.