Research On The Prediction Of Flight Off-block Milestone Time

Flight off-block time is the time for an aircraft to be ready,all doors closed,boarding bridge removed,push back vehicle available and ready to start up immediately upon reception of clearance from the air traffic control tower.It is an important milestone in flight progress monitor since it is able to assist air traffic control to derive pre-departure sequencing in advance,to guide the airport and airlines to carry out ground services as well.Therefore,flight off-block time prediction has gradually become a hot research issue in civil aviation.The existing prediction method takes the empirical statistical method to derive flight off-block time.Its key idea is to obtain the average minimum turnaround time by statistical historical data,and to sum the estimated landing time,taxi-in time and the average minimum turnaround time as the estimated value of flight off-block time.Since the difference of aircraft types,flight schedules and ground support units will cause different turnaround time in large hub airports,the use of a unified turnaround time will have a large deviation and reduce the efficiency of the collaborative operation among airport,airlines and air traffic control.To solve this problem,the study of the flight off-block time prediction is carried out,which uses machine learning methods innovatively instead of statistical historical data to construct a flight off-block time prediction model.Intuitively,a simple multiple linear regression model can be constructed by all pre-order milestones of flight off-block time,but some events that are not relevant or weakly related can interfere with the prediction of the flight off-block time.To solve this problem,a flight off-block time prediction model is proposed based on factor analysis,it calculates the milestones correlation coefficients,and determines the key factors influencing the flight off-block time by calculating factor loading and factor rotation.Then,these key milestone events are as characteristic variables to construct multiple linear regression model.The experimental results show that the average prediction accuracy of the proposed model can reach above 70% and 90% in the error range ±5 and ±10 minutes respectively,which is higher than that of A-CDM empirical method,the method based on principal component analysis and support vector regression.Due to the flight off-block time prediction model based on factor analysis is essentially a multiple linear regression problem via L1-regularization.This model only considers the impact of the pre-order milestones of flight off-block time,ignoring the indirect effects of many factors such as the gate,weather,the timing of the vehicle and whether it is at busy time.It cannot predict flight off-block time more accurately.Because these factors are difficult to quantify,they can be modelled as hidden variables,a flight off-block time prediction model is proposed based on hidden variables.In the training phase,since the optimal target of maximizing data likelihood probability coupling to model parameters,the traditional gradient ascent algorithm cannot be used directly.So the variational expectation maximization algorithm is used to solve the model parameters,which includes expectation calculation and expectation maximization.The expectation calculation aims to optimize and solve the parameters of approximate distribution,and the expectation maximization is to derive regression prediction model parameters by maximizing likelihood probability.The experimental results on the benchmark data set show that,the model based on hidden variables performs better in terms of prediction accuracy and mean square error than the model based on factor analysis.