Time-series prediction method for departure taxi delay situations at airports

To assist airport managers in understanding the operational dynamics of airport surfaces, this study proposes a temporal prediction method for taxi-out delay situations. The process from aircraft pushback to takeoff is divided into three phases: pushback, taxiing, and waiting for takeoff. Considering the surface structure, 20th percentile method (P20) is used to calculate unimpeded taxi out time and unimpeded pushback time. An innovative cumulative taxi delay feature is proposed, and a dataset of surface traffic characteristics is constructed, incorporating factors such as aircraft type, frequency of aircraft operations, and departure rates. K-means clustering is employed to analyze the surface traffic characteristics dataset, allowing for the classification of departure taxi delay situation levels. Finally, a GRU-SVR method is utilized to predict the temporal aspects of taxi-out delay levels. The experimental results show that the deep learning model performs better than the traditional machine learning model in predicting the cumulative taxi delay time, while the traditional machine learning model has a smaller error in predicting the cumulative number of aircraft in taxi delay. After combining with the K-means clustering method, the accuracy of predicting the level of departure taxi delay situations reaches 81.64% under real sample verification, and the accuracy of ±1 level is 99.92%. Therefore, the method proposed in this paper has achieved high accuracy in predicting departure taxi delay situations.