Abstract

International audience; In past decades, airport ground operations have attracted researchers, with the aim of increasing airport efficiency and reducing the environmental impact of airport operations. Airplane taxi operations have received particular attention for their significant impact on the airport efficiency and pollutant emissions and on the fuel cost for airlines. Alternative solutions have been proposed to the engine-on taxi procedures, including the employment of autonomous vehicles to tow the aircraft between gates and runways. In order to be performed, autonomous taxi procedures require precise planning and scheduling by means of sophisticated management systems. At the base of these management systems, lie algorithms for the solution of the routing problem, which provide feasible paths on the airport surface. Two different approaches can be used: compute the paths on the fly, or pre-compute all the possible paths between all the pairs of starting/ending points on the airport grid and store them in a database that is called when needed. In this paper, four different algorithms are implemented and compared for the computation of paths on the fly: two Hopfield-type neural networks and two algorithms based on graph theory. Furthermore, two algorithms for the generation of the path database are presented: a modified version of the Breadth-first search and an implementation of the k-shortest paths algorithm. Each taxi mission, performed by the tractors, consists of three different events, called phases: one central towing phase, where the tractor tows the aircraft between gate and runway and two repositioning phases in which the tractors move from its actual position to the airplane or from the airplane back to the depot.

Abstract

urate prediction of taxi-out time is essential for enhancing airport performance and flight efficiency. In this paper, we apply machine learning techniques to predict the taxi- out time of departure aircraft at Shanghai Pudong International Airport. The exploration of historical data reveals several relevant influencing factors of taxi-out time as well as their correlations. We formulate an extensive system of predictors for our machine learning approach, based on a macroscopic network topology from an aggregate view. The predictors can be divided into 4 categories; namely surface instantaneous flow indices (SIFIs), surface cumulative flow indices (SCFIs), aircraft queue length indices (AQLIs) and slot resource demand indices (SRDIs). Three machine learning methods: linear regression (LR), support vector machines (SVM) and random forest (RF) are formulated using one-day and one-month training samples, and applied to new test dataset to validate the prediction performance. Computational results show that the training RF model using one-month sample significantly outperform other models in terms of prediction accuracy. The proposed methodology can bring significant benefits to analyzing airport ground movement performance and support the activities of airport decision making.