大规模路网下基于多源数据的动态OD矩阵估计理论研究及应用

动态OD矩阵作为动态交通管理系统中的关键基础数据，其准确性直接影响后续交通措施的有效性。传统的动态OD矩阵估计方法主要基于以路段流量为主的单一种类数据，面向中小规模路网进行研究。然而我国城市路网规模宏大且交通数据种类丰富独特，因此如何利用多源交通数据在大规模路网下估计动态OD矩阵具有较高的理论研究价值及现实意义。本项目基于路段流量、监控视频、GPS轨迹、手机信令等多源交通数据，首先提出基于计算图模型的多源数据融合新方法；然后基于双队列模型构建新型动态OD估计理论模型；之后对该模型从时空层面进行分割，利用交替方向乘子法与并行计算技术开发求解算法；最后开发动态OD估计软件，利用深圳、香港两地多源交通数据进行实例验证，并探索在交通管理方面的应用。研究成果能够进一步深化和完善动态OD矩阵估计理论及算法，且有助于分析交通需求的时空分布，为交通管理系统贡献精准基础数据，为交通管理部门提供实时决策支持。

Dynamic Origin-Destination demand Estimation (DODE) plays an essential role in dynamic traffic management systems, and the effectiveness of the management systems largely depends on the precision of the estimated OD demand. Traditional DODE approaches rely on single-source data and most of the literature focuses on small and medium-scale networks. However, road networks in China are large-scale, and multiple sources of data have been collected and archived in many major cities in China. Therefore, there is a great need to study the DODE problem with multi-source data on large-scale networks. Based on the multi-source traffic data, including but are not limited to, traffic volume, traffic surveillance videos, vehicle GPS trajectories, cell phone location data, this project first develops a computational graph-based approach to fuse the data; secondly, a novel DODE formulation based on the double queue model is developed with rigorous analytical properties; thirdly, through spatio-temporal partition of the proposed model, the alternating direction method of multipliers (ADMM) is adopted and extended to make use of the parallel computing technologies for solving the DODE problems on large-scale networks; lastly, an open-sourced DODE software will be developed and applied to Shenzhen and Hong Kong to demonstrate the effectiveness of the proposed approach. The research outcomes could improve the theory and solution algorithm of the DODE problem, and the estimation results can help to analyze the spatio-temporal patterns of dynamic OD demand, support the dynamic traffic management systems, and provide real-time decision support for public agencies.