报告人：邵虎 教授
报告题目：Transportation Network Modeling: Traffic Flow Estimation and Prediction
报告时间：2026年5月12日（周二）下午4：00
报告地点：云龙校区6号楼318会议室
主办单位：数学与统计学院、数学研究院、科学技术研究院
报告人简介：
邵虎，中国矿业大学数学学院，教授，博士，博士生导师，中国矿业大学校学术委员会常委、江苏省应用数学(中国矿业大学)中心副主任、中国矿业大学数学学科建设与指导委员会主任、数学学院教授委员会主任。全国煤炭行业教学名师、全国大学生数学建模竞赛优秀指导教师江苏省高校优秀共产党员，江苏省“青蓝工程”优秀教学团队带头人，江苏省运筹学会副理事长。作为主持人，连续主持5项国家自然科学基金项目(面上4项，青年1项)，主持全国教育规划重点项目1项，省教改项目3项（含重点项目2项），发表科研论文70余篇，出版第一作者专著1部，主编、参编教材4部，获得江苏省教学成果一等奖、教育部自然科学奖二等奖、江苏省教学创新大赛特等奖、中国矿业大学教学贡献奖、教学模范等100余项奖励。主要从事问题驱动型“应用数学”研究，研究方向涉及最优化理论应用、交通网络建模与算法设计、数据驱动下的网络建模与算法、机器学习的应用等。
报告摘要：
Traffic demand flow estimation (TDFE) is a critical task in urban transportation planning and management, as it provides a scientific foundation for decision-making in infrastructure construction, public transit optimization, and congestion mitigation. To address this problem, this study comprises three systematic and interconnected research components: (1) the establishment of an observability theory based on graph isomorphism to guide cost-effective data acquisition, (2) the development of a deep learning-based model for accurate and interpretable dynamic TDFE, and (3) the construction of a bilevel origin-destination (OD) demand model along with the design of a corresponding efficient solution algorithm. To overcome the economic constraints of limited sensing resources, this research first establishes a graph theory-based analytical framework for traffic network flow observability. It derives the analytical relationship for the minimum number of observable links and subsequently proposes a resource-constrained sensor deployment optimization model. By quantifying the information loss imposed by budget limitations, the model aims to maximize network flow observability at minimal cost, thereby laying a high-quality data foundation for subsequent analysis. For accurate and interpretable estimation of dynamic traffic demand, a Multi-feature Recurrent Learning Network (MRLN) that integrates physical traffic mechanisms is developed. This model structures key system processes such as trip distribution, route choice, and traffic assignment into interpretable computational units. Through a temporal-recursive and feature-fusion architecture, it achieves high-fidelity inversion of dynamic origin-destination (OD) demand, significantly enhancing both model interpretability and estimation accuracy. Addressing the practical challenge of solving complex models with real-world, multi-source heterogeneous data, a bilevel OD estimation model is constructed. A tailored, efficient heuristic solving algorithm based on the proximal linearized Alternating Direction Method of Multipliers (ADMM) is designed for this model. This algorithm substantially improves computational efficiency and numerical stability in data-sparse and heterogeneous scenarios, ensuring the practical utility of advanced estimation models. The three research components follow a progressive logic: the economical deployment establishes the essential data foundation for precise modeling, which in turn drives the development of robust solving algorithms. This complete technical chain ensures the practical applicability of advanced models, ultimately delivering an implementable framework for TDFE that provides tangible support for decision-making in intelligent transportation systems.