基于强化学习方法的航班滑出时间预测研究

Research on Flight Taxi out Time Prediction Based on Reinforcement Learning Method

航班的滑出时间是描述机场场面运行状态和周转效率的关键指标,其不确定性会降低航班到达目的机场的可预见性,进而带来航空资源的低效利用和燃油耗费问题。研究了一种基于强化学习的航班滑出时间预测模型。从交通状态和时序特性方面分析并提取影响滑出时间的主要特征集;利用马尔科夫决策过程建模滑出时间预测问题,并通过强化学习算法进行模型训练和测试。在真实机场场面运行数据中进行的实验表明,所提出方法不仅能够准确预测单个航班的滑出时间,还能够捕捉机场场面整体的滑行态势的变化情况,为智慧机场的建设提供新思路。

Flight taxi-out time is a key indicator to describe the operation state and turnover efficiency of the airport surface,and its uncertainty will reduce the predictability of the flight arriving at the destination airport,which will lead to inefficient use of aviation resources and fuel consumption problems.This paper studies a prediction model of flight taxi-out time based on reinforcement learning.First,the main feature set that affects taxi-out time is analyzed and extracted in terms of traffic state and timing characterization.Next,the taxi-out time prediction problem is modeled using a Markov decision process,and the model is trained and tested by the reinforcement learning algorithm.Experiments on real airport surface operation data show that the proposed method can not only accurately predict taxi-out time of a single flight,but also capture changes in the overall taxiing situation of the airport surface,which can provide new insights for the construction of smart airports.