Theories of Urban Traffic Dynamics and Adaptive Control for the Age of Big Data

该项目将探索如何将大数据技术应用于交通信号控制。该项目将利用图论和数据科学的最新进展来识别不同的拥堵模式，并开发独特的治疗方法，以适应城市街道上的交通拥堵。在理想条件下进行了广泛的初步工作，本研究将实现从理想条件到复杂现实环境的飞跃，对减少拥堵和促进国家经济增长产生重大影响。为这个项目设计的算法将是开源的，有可能在普通的交通和通信基础设施上实施。开发一个平台来模拟在现实环境中开发的算法将是一个有价值的教学和推广工具。该项目的变革品质在于其计划：(i)检测和区分交通拥堵可能在城市街道上表现出的各种几何图案；（ii）针对这些不同的模式进行针对性的治疗。几何模式的检测将使用大数据监测技术，并通过将街道网络表示为图形来实现。对这些不同几何图案的处理设想需要重新调整交通信号的时间，既要自适应地测量警戒线社区，又要同步绿色相位，以配合下游队列的消散。这些措施将与另一项措施结合起来，根据需要加快公共汽车通过交通信号的速度。针对现有的社区交通模型，以及无模型的强化学习方法，设想了特定的增强功能。改进后的方法将用于预测治疗在复杂的现实环境中的表现。该项目还将探索如何利用大数据技术测量和处理的输入，不断改进模型和无模型预测。进一步的探索将是将精确的预测与优化技术结合起来，这样处理方法就可以在一天中适应城市不断变化的拥堵模式。一旦以上述方式推广，将使用模拟对处理进行测试和改进。然后将设计上述算法，并将改进的处理方法编码到模拟平台中，用于教学和推广目的。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The project will explore how big-data technologies can be applied to traffic signal control. The project will leverage recent advances in graph theory and data science to identify distinct congestion patterns and develop unique treatments that would adaptively treat traffic congestion on urban streets. Extensive preliminary work has been conducted under idealized conditions and this research will make the leap from idealized conditions to complex, real-world settings, for significant impact in reducing congestion and aiding economic growth for the nation. Algorithms designed for this project will be open-source, with the potential to be implemented on ordinary traffic and communication infrastructures. The development of a platform to simulate the developed algorithms in realistic settings will be a valuable teaching and outreach tool. The project's transformative quality lies in its plans to: (i) detect and distinguish the various geometric patterns that traffic congestion can exhibit on city streets; and (ii) target treatments to suit those distinct patterns. The detection of geometric patterns will be pursued using big-data monitoring technologies, and by representing street networks as graphs. Treatments envisioned for those distinct geometric patterns entail the re-timing of traffic signals, both to adaptively meter cordoned neighborhoods and to synchronize green phases to coincide with the dissipation of queues immediately downstream. These will be integrated with another treatment that expedites bus movements past traffic signals on as-needed bases. Specific enhancements are envisioned for existing models of neighborhood traffic, and for model-free, Reinforcement Learning approaches. The enhanced methods will be used to predict how the treatments perform in complex, real-world settings. The project will also explore how model and model-free predictions might be continually refined using inputs measured and processed with big-data technologies. Further exploration will then go toward coupling the refined predictions with optimization techniques, so that treatments can adapt over the day to suit a city's evolving congestion patterns. Once generalized in the above ways, the treatments will be tested and refined using simulation. The aforementioned algorithms will then be designed, and the refined treatments will be coded into the simulation platform for teaching and outreach purposes.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.