Encoding Dynamic Traffic Flow Analysis into AI for Network-Wide Early Alarming of Traffic-Demand-Influencing Events and Their Impacts

将动态交通流分析编码到人工智能中，以便对影响交通需求的事件及其影响进行全网早期预警

• 批准号: 2213459
• 负责人: Lili Du
• 金额: $ 30万
• 依托单位: University of Florida
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2025-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2213459&HistoricalAwards=false
• 关键词: Encoding; Dynamic; Traffic; Flow; Analysis

This project will integrate dynamic traffic flow analysis into artificial intelligence to provide early alarming of significant traffic-demand-influencing events (DIEs) and the associated traffic impacts. Urban traffic can deviate from normal states due to various scheduled and unscheduled DIEs, such as sports, commercial promotions, and festivals. These events often induce a surge in traffic demand, cause hours of congestion, and affect multiple traffic infrastructures. Early awareness of DIEs and their traffic impact will benefit many stakeholders, including travelers, government, and transportation-related service providers, in taking proactive actions to manage traffic congestion. This project aims to develop a network-wide online DIE monitoring system, which can automatically provide early alarming of the DIEs and forecast the resulting congestions. The research outcomes can be directly employed to mitigate traffic network congestions and become an essential component of future smart city technologies. The interdisciplinary studies will open a new line of research on seamlessly integrating transportation engineering, data science, and artificial intelligence-based technologies to develop new scientific knowledge and methodologies for traffic operations and control. The PIs will integrate research into pedagogical developments at their home departments, actively disseminate research results, and engage in K-12 outreach activities via the Gator Outreach program at the University of Florida.This project will develop hybrid approaches that integrate traffic flow theories, optimization algorithms, high-dimensional machine learning, and artificial intelligence for monitoring the events that cause significant traffic demand surges in a (sub)urban area based on real-time temporal-spatial traffic data. The research will produce the following transformative scientific technologies: (1) Data science-empowered online shockwave generating algorithms to accommodate data imputation, which quantitatively captures the impact of a DIE on traffic conditions on a network over time; (2) innovative encoding approaches to compile shockwave diagrams for feeding machine learning model better; (3) sparse principal component analysis-based feature engineering for selecting and fusing the most promising subset of features for DIE monitoring; (4) a novel feature acquisition location recommendation scheme to determine where new features should be additionally acquired from the traffic network to boost the DIE-monitoring AI maximally; and (5) a radical extension to the recurrent neural network (RNN) model and theory by incorporating explicit regularization and distributed computing-compatible architectures.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.

该项目将把动态交通流分析集成到人工智能中，以提供重大交通需求影响事件（DIE）和相关交通影响的早期预警。城市交通可能会因各种预定和非预定的DIE（如体育、商业促销和节日）而偏离正常状态。这些事件通常会导致交通需求激增，造成数小时的拥堵，并影响多个交通基础设施。及早了解DIE及其对交通的影响将使许多利益相关者受益，包括旅行者、政府和交通相关服务提供商，从而采取积极行动来管理交通拥堵。本项目旨在开发一个全网络的在线DIE监测系统，该系统可以自动提供DIE的早期报警并预测由此产生的故障。其研究成果可直接用于缓解交通网络拥堵，并成为未来智慧城市技术的重要组成部分。跨学科研究将开辟一条新的研究路线，无缝集成交通工程，数据科学和基于人工智能的技术，为交通运营和控制开发新的科学知识和方法。 PI将把研究融入到他们所在部门的教学发展中，积极传播研究成果，并通过佛罗里达大学的Gator Outreach计划参与K-12推广活动。该项目将开发混合方法，将交通流理论，优化算法，高维机器学习，以及人工智能，用于基于实时时间-空间交通数据来监测在（次）城市区域中引起显著交通需求激增的事件。该研究将产生以下变革性的科学技术：（1）数据科学授权的在线冲击波生成算法，以适应数据插补，该算法定量捕获DIE随时间推移对网络交通状况的影响;（2）创新的编码方法，以编译冲击波图，从而更好地为机器学习模型提供支持;（3）基于稀疏主成分分析的特征工程，用于选择和融合用于DIE监测的最有希望的特征子集;（四）一种新颖的特征获取位置推荐方案，用于确定应该从交通网络的何处额外获取新特征，最大限度地提高了芯片监控AI;（5）通过结合显式正则化和分布式计算兼容架构，对递归神经网络（RNN）模型和理论进行了根本性的扩展。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

A Traffic Flow Dependency and Dynamics based Deep Learning Aided Approach for Network-Wide Traffic Speed Propagation Prediction

• DOI: 10.1016/j.trb.2022.11.009
• 发表时间: 2023-01
• 期刊: Transportation Research Part B: Methodological
• 作者: Hanyi Yang;Lili Du;Guohui Zhang;Tianwei Ma