基于Q-learning算法的机场航班延误预测  

Airport Flight Delay Prediction Based on Q-learning Algorithm

作　　者：刘琪[1] 乐美龙 LIU Qi;LE Mei-Long(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China)

机构地区：[1]南京航空航天大学,江苏南京211000

出　　处：《航空计算技术》2025年第1期28-32,共5页Aeronautical Computing Technique

基　　金：国家自然科学基金项目资助(71471110);南京航空航天大学研究生创新基地开放基金项目资助(xcxjh20220721)。

摘　　要：将改进的深度信念网络(DBN)和Q-learning算法结合建立组合预测模型。首先将延误预测问题建模为一个标准的马尔可夫决策过程,使用改进的深度信念网络来选择关键特征。经深度信念网络分析,从46个特征变量中选择出27个关键特征类别作为延误时间的最终解释变量输入Q-learning算法中,从而实现对航班延误的实时预测。使用北京首都国际机场航班数据进行测试实验,实验结果表明,所提出的模型可以有效预测航班延误,平均误差为4.05 min。将提出的组合算法性能与4种基准方法进行比较,基于DBN的Q-learning算法的延误预测准确性高于另外四种算法,具有较高的预测精度。This article combines an improved Deep Belief Network(DBN)with Q-learning algorithm to establish a combined prediction model.Firstly,model the delay prediction problem as a standard Markov decision process,using an improved deep belief network to select key features.Through deep belief network analysis,27 key feature categories were selected from 46 feature variables as the final explanatory variables for delay time and input into the Q-learning algorithm to achieve real-time prediction of flight delays.The test experiment was conducted using flight data from Beijing Capital International Airport,and the results showed that the proposed model can effectively predict flight delays with an average error of 4.05 minutes.Comparing the performance of the proposed combination algorithm with four benchmark methods,the delay prediction accuracy of the DBN based Q-learning algorithm is higher than the other four algorithms,indicating a higher prediction accuracy.

关 键 词：航空运输 航班延误预测 深度信念网络 Q-LEARNING 航班延误 

分 类 号：V35[航空宇航科学与技术—人机与环境工程]