Strategic air traffic flow management under uncertainties using scalable sampling-based dynamic programming and Q-learning approaches
Strategic air traffic flow management under uncertainties using scalable sampling-based dynamic programming and Q-learning approaches
复制标题
使用可扩展的基于采样的动态规划和 Q 学习方法在不确定性下进行战略空中交通流量管理
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
F. Lewis
中科院分区：
文献类型：
--
作者：
Junfei Xie;Yan Wan;F. Lewis
许多信息物理系统（CPS）呈现出高维度的环境不确定性，这种不确定性会调节物理系统的动态，并使计算决策任务变得复杂。例如，战略空中交通流量管理（ATFM）旨在通过对较长前瞻时间的交通流量进行管理来解决空中交通拥堵问题。对流天气作为交通拥堵的一个主要因素，具有很大的不确定性，这使决策过程变得极为复杂。其他不确定性，比如交通需求，进一步扩大了不确定性空间，并使对不确定性具有鲁棒性的最优控制解决方案的设计变得复杂。在本文中，我们将战略空中交通流量管理表述为一个随机最优控制问题，并使用基于可扩展采样的动态规划和Q学习方法来应对高维度不确定性。仿真研究证明了这些方法的有效性。这些方法通常适用于在高维度环境不确定性下运行的信息物理系统。
Many Cyber-Physical Systems (CPSs) demonstrate high-dimensional environmental uncertainties that modulate the physical system dynamics and complicate the computational decision-making tasks. As an example, strategic air traffic flow management (ATFM) aims to resolve air traffic congestion through managing traffic flows at a long look-ahead time. The convective weather, a dominant factor of traffic congestion, is very uncertain and significantly complicates the decision-making process. Other uncertainties such as traffic demands further expand the uncertainty space and complicate the design of optimal control solutions that are robust to uncertainties. In this paper, we formulate the strategic ATFM as a stochastic optimal control problem, and use scalable sampling based dynamic programming and Q-learning approaches to address high-dimensional uncertainties. Simulation studies demonstrate the effectiveness of these approaches. These approaches generally apply to CPSs that operate under high-dimensional environmental uncertainties.