A data-driven method is proposed to identify regional traffic TOD intervals through analyzing traffic flow data w hich are collected from multi road intersections ,multiple directions and different times .Firstly ,traffic data will be checked by three kinds of screening rules and outlier data will also be detected by visual and statistical methods .Invalid and missing data can be interpolated by several interpolation algorithms ,such as EM and MCMC methods .Secondly ,from the perspective of spatial scale ,critical traffic flows in traffic network are identified by using multiple correlation analysis and principal component analysis ,and four statistical indexes are applied to calculate the number of critical traffic flows .Thirdly ,from the perspective of temporal scale ,traffic states of different time are clas-sified by hierarchical clustering ,and finally TOD intervals are ascertained by using statistical histo-gram .This method is also demonstrated by analyzing nine intersections’ vehicle data at Beijing .Six different critical flow s are identified from 37 traffic flow s ,and five traffic states are also obtained by clustering 15-minute monitoring data ,and eight T OD intervals are finally determined .Each T OD in-terval is a peak ,mid or off-peak period of artery or region traffic .It show s that this data-driven meth-od is effective and practical .%提出了一种基于数据驱动的T OD时段识别方法，对区域不同路口、不同流向、不同时刻的交通流数据，采用多元相关分析、主成分分析等在空间尺度上识别出路网的关键路口和关键交通流向，采用层次聚类在时间尺度上识别出不同的交通状态和各T OD时段。以9个道路交叉口流量数据为应用实例，获取其中6个不同路口方向为关键交通流，并将不同时刻观测值聚类为5种不同的交通状态，进而识别出1 d的8个 T OD时段，每个时段分别代表干线或区域高、中、低等不同流量时期，表明了该方法的有效性。
展开▼
