Collaborative Research: Transportation Network Identification: Information Fusion via Stochastic Optimization

在正常和极端情况下，对于网络监控和有效的交通系统管理和控制来说，交通网络状态的知识是至关重要的，例如，关于网络级别的交通流量和旅行时间的知识。最近的技术进步，如移动传感和联网车辆，可以产生大数据。总的来说，这项研究旨在解决如何最好地利用这些固有的大规模、动态和异质(多源)数据流的挑战。更具体地说，该项目寻求一种创新的系统方法，通过在复杂的网络结构上整合各种交通数据片段(如固定传感器收集的交通计数和蓝牙标签读取器日志、移动电话记录和全球定位系统跟踪收集的个人“城市数字足迹”)，来估计交通流量和旅行时间的统计特性。该项目的成功成果将通过更有效地利用信息，从而更可持续和更有效地规划和运营运输系统，直接造福社会。该项目包括课程开发和学生指导活动，以帮助下一代运输专业人员更好地为大数据时代带来的挑战做好准备。从数学上讲，该项目涉及的问题是：给定可直接测量的网络参数x(往往是本地化和不完整的)，如何通过建立在复杂网络结构上的y和x之间的映射来推断通常难以直接测量的全球网络参数y？这类问题在交通、通信和能源网络中有着广泛的应用；不过，本项目的重点是交通网络。基于交通网络科学、随机优化、变分分析和非参数估计的知识，研究小组将从事以下主要工作：(1)建立基于随机优化和非参数估计的网络识别优化框架；(2)通过约束和功能映射将多源交通数据(硬信息)与基于领域知识的软信息相结合；(3)通过贝叶斯方法将基于历史数据的离线估计与实时信息联系起来，用于在线估计和决策支持；(4)使用真实数据和计算机模拟来测试和验证项目的方法。本研究将交通网络科学知识与数据分析相结合，建立了统一的交通网络识别理论框架。这种新方法在处理各种类型的硬信息和软信息以及捕捉交通网络物理信息方面提供了比现有方法更大的建模灵活性，从而带来了交通网络系统辨识的范式转变。该项目还为加强优化和统计之间的联系创造了一个现实世界的工程平台。

Knowledge of the traffic network state, in terms of, for example, network-level traffic flow and travel time, is critically needed for network monitoring and effective transportation system management and control under both normal and extreme conditions. Recent technological advances, such as mobile sensing and connected vehicles, can generate big data. In general, this research aims to address challenges on how to best use these inherently large-scale, dynamic, and heterogeneous (multi-source) data streams. More specifically, the project seeks an innovative systems approach for estimating the statistical properties of traffic flow and travel time via integrating various traffic data pieces over a complex network structure (such as traffic counts collected by fixed sensors and individual "urban digital footprints" collected by Bluetooth tag reader logs, cellular phone records, and global positioning system traces). A successful outcome of this project will directly benefit society through more effective utilization of information, and thus more sustainable and efficient transportation system planning and operations. This project includes curricula development and student mentoring activities that help better prepare next-generation transportation professionals for challenges brought by the big data era.Mathematically, the question addressed in this project is: Given directly measurable network parameters x, which tend to be localized and incomplete, how can one infer global network parameters y that are often difficult to be measured directly, with the mapping between y and x built on a complex network structure? This problem category has broad applications in transportation, communication, and energy networks; although, the focus in this project is on transportation networks. Built on knowledge in transportation network science, stochastic optimization, variational analysis, and non-parametric estimation, the research team will pursue the following main tasks: (1) Creation of an optimization framework for network identification based on stochastic optimization and non-parametric estimation; (2) Integration of multi-source traffic data (hard information) with domain-knowledge-based soft information via constraints and functional mapping; (3) Linking historical-data-based offline estimation with real-time information for online estimation and decision support through Bayesian methods; (4) Testing and validating the project's methods using both real-world data and computer simulations. This research establishes a unifying theoretical framework for traffic network identification that integrates knowledge in transportation network science and data analytics. By providing greater modeling flexibility than existing methods in handling various types of hard and soft information and in capturing transportation network physics, this new method ushers a paradigm-shift in traffic network system identification. The project also creates a real-world engineering platform for strengthening connections between optimization and statistics.