进港航班排序强化学习模型研究

发布时间：2018-03-15 04:06

本文选题：智能体　切入点：空中交通管制　出处：《工程科学与技术》2017年S2期 　论文类型：期刊论文

【摘要】：为解决进港航班排序中智能化程度不高的现实问题,提出进港航班排序强化学习模型。首先,确定进港航班排序强化学习模型的状态、动作、智能体、环境、奖赏函数、约束条件、Q学习等。进港航班排序强化模型中的状态是各进港航班的到达时刻,动作是对航班到达时间的调整,智能体对航班的到达时刻进行调整,环境对动作做出反应,一个新的到达时间和奖赏值被传给智能体,奖赏函数考虑了延误时间、经济成本、对后续航班的影响。该模型考虑了航班不能提前降落,分配的到达时间不早于计划的到达时间,进港航班流量不能超过机场的到达容量值等约束条件。使用双流机场进港航班数据对该模型进行验证。对比分析先到先服务和强化学习模型的排序、延误时间、延误成本、后续航班延误成本和奖赏值。先到先服务算法的奖赏函数值为3 164,强化学习算法的奖赏函数值为2 880,强化学习模型更优。模型中奖惩函数的评价指标、权重、约束条件可以根据管制工作实际情况进行设置,该模型可以为空中交通管制人员进行进港航班排序提供决策支持。
[Abstract]:In order to solve the problems in real flight sequencing in the intelligent degree is not high, the inbound flights sort of reinforcement learning model. First, determine the inbound flights sort of reinforcement learning model of the state, action, agent, environment, reward function, constraint condition, Q learning. In class ranking model in strengthening port state is the entrance the arrival time of the flight, action is on flight arrival time adjustment agent on the flight arrival time is adjusted in response to environmental action, a new arrival time and the reward value is passed to the agent, the reward function considering the delay time, the economic cost, impact on the subsequent flights of the model. The flight cannot advance landing, distribution of the arrival time is not earlier than the planned arrival time, flight arrival flow cannot exceed the airport arrival capacity value as constraint conditions. The use of Shuangliu Airport inbound flight data To verify the model. A comparative analysis of first come first serve and the reinforcement learning model of sorting, delay time, delay cost, subsequent flight delay cost and reward value. First come first serve algorithm of the reward function value is 3164, a value of 2880 reward reinforcement learning algorithm, reinforcement learning model is better. The weights of evaluation indexes model, reward and punishment function, constraints can be set according to the actual situation of control, the model for air traffic control personnel arrival sequencing and scheduling decision support.

【作者单位】：四川大学视觉合成图形图像技术国防重点学科实验室;
【基金】：国家空管委科研资助项目(GKG201403004)
【分类号】：TP181;V355

【参考文献】