Dynamic Prediction of Ground Progress of Flights Based on Bi-LSTM

Ground operation process perception for transit flights is an important function of airport collaborative decision-making systems. Currently, there are limitations in achieving refined process prediction. Furthermore, the reliability and accuracy of this prediction are limited. To overcome this challenge, a dynamic prediction method for ground operation processes of transit flights based on Bi-LSTM is proposed. A ground operation process network model was established, a feature extraction method based on the spatio-temporal relationship of nodes was designed, and a dynamic prediction method for ground operation process was constructed by combining the attention mechanism. Simulation results based on actual data from a hub airport show that the proposed method can ensure dynamic prediction at each node. It also fully considered the relationships between individual nodes. The method has an average absolute error of only 1.8 min, which is up to 10 min higher than that of other models. Compared with other models, the average absolute error of the prediction results improves by up to 10 min. It can provide an objective decision-making basis for the short-term tactical organization of airport operations.