机场航班延误恢复的强化学习算法  

Reinforcement Learning Algorithm for Airport Flight Delay Recovery

作　　者：丁建立[1] 刘德康 DING Jianli;LIU Dekang(School of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300

出　　处：《重庆交通大学学报（自然科学版）》2024年第9期50-58,共9页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：国家自然基金民航联合研究基金重点支持项目(U2033205,U2233214)。

摘　　要：机场出现航班延误会导致飞行器和乘客滞留机场,若航班延误恢复调度不当会扩大延误造成的损失。针对航班延误恢复调度的损失最小化问题,设计了延误总损失计算的目标函数,构建航班延误恢复马尔科夫决策过程,建立了机场航班延误恢复重排班模型。为了解决计算的复杂性问题,采用深度学习神经网络参数化策略函数对减小延误损失目标函数值的策略进行参数化,利用奖励函数和优势函数对其进行训练,提出了一种机场航班延误恢复强化学习算法。研究结果表明:该算法能够将航班延误总损失降低7.83%,将旅客延误时长降低7.23%,相比于其他算法,该算法在时间和性能上均取得优势。Flight delays at airports resulted in aircraft and passengers being stranded at the airport,and improper recovery and scheduling of flight delays can exacerbate the losses caused by delays.Aiming to the issue of minimizing losses in flight delay recovery scheduling,a target function was formulated to calculate the total delay loss,a Markov decision-making process was constructed for flight delay recovery,and an airport flight delay recovery rescheduling model was established.To address computational complexity,a deep learning neural network parameterized policy function was employed to parameterize the strategy of reducing the delay loss objective function value,which was trained by the reward function and advantage function.A reinforcement learning algorithm for airport flight delay recovery was proposed.The research results show that the proposed model can reduce the total loss of flight delays by 7.83%and the duration of passenger delays by 7.23%.The proposed deep reinforcement learning algorithm outperforms other algorithms in both time and performance.

关 键 词：交通运输工程 航班延误恢复 延误损失 航班重排班 马尔科夫决策 深度强化学习 

分 类 号：U8[交通运输工程] TP399[自动化与计算机技术—计算机应用技术]