Research and comparison on identification and prediction methods of air traffic network congestion

Air traffic congestion (ATC) identification and prediction is a prerequisite work for air traffic flow management (ATFM) and flight delay mitigation. In this paper, we construct the ex-post evaluation index system of air traffic network (ATN) congestion and propose the ATN congestion identification model based on fuzzy C-means clustering (FCM). According to the results of congestion levels identification, this paper analyzes various factors affecting ATN congestion and constructs the ex-ante prediction index system of ATN congestion. The ATN congestion levels prediction is performed by decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) algorithms. The grid search method is used to adjust and optimize the key parameters of models. The empirical results of ATN congestion levels prediction in Northeast China show that the accuracy of XGBoost and LightGBM algorithms for ATN congestion levels prediction respectively reaches 86.4% and 86.6%, which are better than mainstream classification prediction algorithms (DT and RF). With comparable prediction accuracy, LightGBM outperforms XGBoost in computational speed and stability of prediction for different congestion levels.