A Data-driven Optimization Method For Flight Pushback Scheduling

Optimizing airport departure process is of vital importance to improve airport operation performance,and the key is to optimize flight pushback schedule.The optimization of pushback schedule is based on the prediction of flight taxi out time,and take departure normality and airport operation efficiency as the optimization target.In view of this,this paper applies a variety of machine learning algorithms to predict flight taxi out time,and further design a data-driven pushback scheduling method.The main innovative work of this paper are listed as follows:(1)Based on the analysis of the operation data from the collaborative decision management system of Beijing Capital International Airport(BCIA),the surface traffic characteristics of BCIA were summarized.After integrated with current research,the basic feature set was built for predicting flight taxi out time.Taking the basic feature set as input,the prediction effect of ten machine learning algorithms on taxi out time was discussed.With the best prediction model,a series of feature analysis methods were used to find the influence of each feature on prediction quality.The results show that Boosting ensemble method,which represented by GBRT and XGBoost has the best overall prediction performance.The co-taxiing activity and runway operation mode affect flight taxi out time most at BCIA.(2)Taking departure normality and airport operation efficiency as objectives,the flight pushback scheduling model was built by integrating the previous taxi out prediction model.Single-objective and bi-objective model solving algorithms were designed respectively.Aiming at the difficulty of model objective function estimation which caused by obtaining prediction features and the error of prediction model,a data-driven feature estimation method was proposed and the flight taxi out time was convert to probability distribution form based on prediction value and error.A reasonable objective estimation method then be designed based on taxi out time distribution.Besides,the randomness brought by feature estimation were handled with sample average approximation.(3)Six representative dataset were selected from the operation data of BCIA for case study.The results show that the proposed method can effectively improve the departure normality by adjusting the expected departure delay flights to the slot with lower surface operating load.The increase of departure traffic will affect the algorithm optimization strategy,intensify the competition between normality and efficiency,and make the two objectives deteriorate simultaneously.Reasonably set the normality assessment standard and adjustment range of pushback schedule according to the demand of departure can ensure the operation performance of airport as well as obtain better management benefit.