关键词: 航空运输, 机场地面保障, 深度强化学习, 保障调度, 条件任务图, 混合整数规划模型, 图卷积网络

Abstract: Airport ground handling covers the key operational links from flight landing to takeoff, and there are conditional temporal dependencies between tasks in each link. At the same time, the multi flight operation process is accompanied by resource competition, and this complex coupling relationship has become a major challenge restricting the development of ground support. Traditional methods rely on manual rule design and are difficult to apply in large- scale and high dynamic scenarios. This paper proposes a deep reinforcement learning algorithm framework that integrates mathematical graph embedding and attention mechanism for the airport ground handling scheduling. The dynamic scheduling problem with random task sequences is graphically modeled as a mixed integer programming model through conditional task graphs, and transformed into a heterogeneous graph. The graph convolutional network is used to extract global graph embedding to efficiently represent the coupling relationship in the graph, which is fused with real-time state to form a deep reinforcement learning state representation. The attention mechanism based policy network is used to learn serialized scheduling strategies, achieving end-to-end scheduling strategy optimization. The test results in a real airport simulation environment show that compared with traditional heuristic methods and reinforcement learning methods, this algorithm framework has significant improvements in multiple core metrics. Among them, compared with deep Q-networks, the average flight delay is reduced by 40.1%, and the resource utilization rate is improved by 6.6%, which verifies the effectiveness and superiority of the algorithm.

Key words: air transportation, airport ground handling, deep reinforcement learning, handling scheduling, conditional task graph; mixed integer programming model, graph convolutional network