CAREER: Probabilistic Network Flow Theory: Embracing Emerging Big Data for Efficient, Reliable and Sustainable Multi-modal Transportation Systems

该学院早期职业发展计划（Career）项目将建立一个基于交通系统建模和融合不同系统（包括道路、公共交通和停车场）的大规模多源数据的数学框架。目标是利用不同尺度的出行需求的时空特征，了解网络中断如何概率地影响交通系统，并帮助促进有关规划和实时运营的决策制定。该项目将为数据驱动的交通科学奠定基础，这将对从个人生活质量到国家经济的各个层面产生影响。该项目将涉及与几家公共机构和私营公司合作，在匹兹堡大都市区开发、部署和测试基于大规模数据分析的真实世界系统。所有模型、算法和示例都将在公共领域实现和开源，以促进对应用程序的访问，并引发来自世界各地用户的讨论。研究结果将用于开发一门新的基础设施管理数据分析本科课程。此外，将与卡内基自然历史博物馆合作开发一个虚拟实验室，用于教育7-12年级的学生、大学生和一般公众。来自卡内基梅隆大学和匹兹堡大学的学生将通过学习课程、数据分析竞赛和实践活动参与其中。本CAREER项目的目标是开发利用大规模数据推断概率网络流特征的理论和算法，并在不确定性下优化管理交通网络。在流量动力学和用户行为的背景下，使用高维联合概率分布来显式地建模概率网络流和系统状态。这些联合概率分布是通过融合和挖掘多年来收集的细粒度数据来学习、估计和预测的。它们揭示了流动和系统状态的二阶统计量，即方差-协方差，从而在高时空粒度下更好地理解流动/系统特征之间的相互关系。该项目还将发展严格的统计理论，通过动态网络划分来识别子网中经常性和非经常性的流量模式。将概率网络流整合到规划和运营的决策理论中，将推动网络优化科学的发展。如果成功的话，这项研究将为以高效、全面、及时和可靠的方式感知、建模、设计、规划和运营复杂的基础设施网络创造一个新的范例。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。

This Faculty Early Career Development Program (CAREER) project will establish a mathematical framework based on transportation system modeling and fusion of large-scale multi-source data across different systems including roadway, public transit and parking. The goal is to exploit the spatio-temporal characteristics of travel demand at different scales, understand how network disruption probabilistically affects the transportation systems, and help facilitate decision making regarding planning and real-time operations. The project will lay the foundation for data-driven transportation science that is to have an impact at all scales from individuals' quality of life to the nation's economy. This project will involve collaboration with several public agencies and private firms to develop, deploy and test real-world systems in the Pittsburgh metropolitan area based on large-scale data analytics. All models, algorithms, and examples will be implemented and open sourced in the public domain to promote access to applications and spark discussions from users all over the world. The results will be used to develop a new undergraduate course on data analytics for infrastructure management. Additionally, a virtual laboratory will be developed in conjunction with Carnegie Museum of Natural History for educating students in grades 7-12, college students, and the general public. Students from both Carnegie Mellon University and University of Pittsburgh will be engaged through learning sessions, data analytics competitions, and hands-on activities. The goal of this CAREER project is to develop theories and algorithms that utilize large-scale data to infer characteristics of probabilistic network flow and optimally manage transportation networks under uncertainty. High-dimensional joint probability distributions are used to explicitly model probabilistic network flow and system states in the context of flow dynamics and user behavior. Those joint probability distributions are learned, estimated, and predicted from fusing and mining fine-grained data collected over many years. They reveal the second-order statistics of flow and system states, namely variance-covariance, resulting in a better understanding of the inter-relations among flow/system characteristics, at high temporal and spatial granularity. This project will also develop rigorous statistical theories to identify recurrent and non-recurrent flow patterns in subnetworks through dynamic network partition. The science of network optimization will be advanced by integrating probabilistic network flow into decision making theories for both planning and operation. If successful, this research creates a new paradigm for sensing, modeling, designing, planning and operating complex infrastructure networks in an efficient, holistic, timely and reliable manner.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.