报告题目：A computational graph-based deep learning model for traffic demand flow estimation
主讲人：邵虎教授（中国矿业大学）
时间：2026年1月14日（周三）17:00 p.m.
地点：北院卓远楼305会议室
主办单位：统计与数学学院
摘要：
This paper introduces a computational graph-based deep learning model, the multi-channel recurrent learning network (MRLN), for addressing the problem of traffic demand flow estimation. The MRLN incorporates two fundamental techniques: horizontal multi-channel learning (MCL) and vertical recurrent learning (RL). From the horizontal view, the MCL learns various traffic states and travelers' routing preferences of by utilizing different sets of paths in different channels. Simultaneously, from the vertical perspective, the RL integrates origin-destination matrix estimation and static traffic assignment into a unified model to capture dynamic temporal dependencies between OD demand flow and link flow. By further aggregating the feature representations learned from both MCL and RL, we are able to estimate the OD demand flows and the link flows for each timestep. In the MRLN, forward propagation is utilized to describe the transmission process of traffic flow from the OD layer to the path layer and then to the link layer, as well as the mapping relationships between previous timesteps and subsequent timesteps. The MRLN model's parameters are updated via backpropagation algorithms. Experiments conducted on three transportation networks comprising the Sioux Falls network and a large-scale Melbourne transportation network verify the performance of the MRLN model. The results demonstrate that introducing multiple channels can effectively reduce the estimation error of traffic flows.
主讲人简介：
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、国家一流课程负责人，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，参编教材2部，获得江苏省教学成果一等奖、教育部自然科学奖二等奖、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。
