Abstract:

The flight departure process is dynamic， changeful and highly random. The uncertainty of the taxi-out time makes it difficult to scientifically allocate the pushback time. In this paper， the problem of flexible control of aircraft pushback time is studied based on probabilistic distribution of taxiing time. Firstly， a machine learning-based probabilistic distribution of taxiing time prediction model， a hyper parameter tuning strategy and a probabilistic prediction performance evaluation indicator system are established using random forests regression together with the kernel density estimation. Then， the concept of “buffer” is introduced， and a flexible control method of “pushback time” for departure flight is proposed. The effect of buffer size on the probability of reaching the runway head on time is clarified. Finally， by applying the chance-constrained theory of the stochastic programming， a flexible control method for the “pushback slot” of the departure flight is proposed， and the definition rule for the feasible pushback time range that satisfies change constraints is designed. Experimental results show that the probabilistic taxiing time of the departure flight can be predicted scientifically with the proposed prediction method， and flexible transformation from traditional rigid control to multi-view flexible control of the pushback time is then realized. Our study can provide support for enhancing the predictability and flexibility of airport departure control.

Key words: departure/pushback time control, taxiing time prediction, taxiing buffer time, random forest regression, probability density function, kernel density estimation, chance-constrained condition