数据不完备条件下基于个体行为的大规模公交网络出行矩阵估计研究

Giving priority to public transportation system is a must to alleviate urban congestion. Fundamental transit data provide reliable support for transit operation in an effective and scientific manner. Transit smart card data and GPS data can be used to obtain transit OD matrices. However, existing studies are either based on survey or small-scale data with good quality, and fail to take into account both statistical dependency for various transit data sources and travel behavior heterogeneity for different travelers. This research project focuses on individual-level transit OD matrix estimation model with incomplete data sets. By fusing both transit smart card data and GPS data, the spatiotemporal trajectories for transit vehicles can be reconstructed. A Bayesian decision tree based model is proposed to infer passenger boarding stops with missing GPS data; By mining multi-day transit smart card transaction records, a non-hierarchical spatial clustering algorithm is developed to extract travelers’ historical travel patterns and features. Based on spatial and temporal constraints for transfer activities, passenger alighting stops can be estimated. Finally, individual trip chain can be constructed. Considering the huge data size, transit OD matrix optimization techniques for large-scale transit network are adopted. These techniques are based on statistical learning theory. The outcomes of this research project can be used for urban transit route planning and real-time transit performance monitoring.

优化发展公共交通系统是我国缓解城市交通拥堵的必然选择，公交基础数据为有效、科学地进行公交运营决策提供了可靠的基础保障，公交IC卡数据以及GPS数据可以用于获取公交出行矩阵，已有的研究多采用人工调查方式或对数据质量较好的线路进行宏观估计公交出行量，未充分考虑多源公交基础数据间的统计关联性与出行者个体行为差异性。本项目重点研究基于不完备数据集的个体公交出行矩阵估计模型：通过IC卡数据与GPS数据融合算法，重建车辆时空行驶轨迹；在GPS数据缺失条件下，研究基于贝叶斯决策树的上车站点推算模型；通过对多天IC卡出行记录的挖掘，拟采用空间非层次聚类算法获取出行者历史出行模式与特征，结合换乘时空约束条件，推算出行者下车站点，最终形成乘客个体出行链；考虑到庞大的数据规模，研究基于统计学习理论的大规模公交网络出行矩阵优化算法。本项目成果可为城市公交线网规划、公交实时运营监控提供可靠的支持与保障。

优化发展公共交通系统是我国缓解城市交通拥堵的必然选择，公交基础数据为有效、科学地进行公交运营决策提供了可靠的基础保障，公交IC卡数据以及GPS数据可以用于获取公交出行矩阵。本项目重点研究基于不完备数据集的个体公交出行矩阵估计模型，识别通勤乘客职住时空分布；在公共运营方面，考虑公交上下车行为，对公交串车现象进行预测，并对乘客候车时间范围进行估计；在网络层面，提出多尺度径向基函数算法预测极端情况地铁客流突变，并开发基于E科学的公交网络性能评价系统。研究成果为公交线网优化提供可靠的数据支持，有助于提升公交运营管理合理性，提高公交服务水平，可广泛应用于城市公共交通规划的决策及技术部门。