DescriptionThe emergence of Transportation Big Data provides rich information for estimating and predicting urban travel demand patterns. The traditional travel demand sensors involve labor-intensive survey data, traffic detector data for assignment model calibration, vehicle re-identification data from scattered Bluetooth, Wifi, or License plate readers, or aggregated cellphone activity data used in existing dynamic Origin-Destination estimation models or applications. With the growing number of mobile devices with GPS units and improvement in WLT technologies, the Location-Based Social Network (LBSN) data is an emerging travel demand data source. LBSN data recorded check-in or tweeting activities of massive users at different points of interests (POIs). The wide-range of POIs ensure the dense coverage of the main urban areas and the user-confirmed POI information provides the much-needed trip purpose information not available in other data sources. Meanwhile, the LBSN data has the advantages of passive secondary data collection usually not for the purpose of travel surveys, and anonymization. Despite the above advantages, LBSN data is not without its limitations for estimating urban travel demand, as well as dynamic OD estimation for proactive urban congestion mitigation and operations. First, LBSN can have a systematic temporal error for estimating travel demand. The LBSN activities do not always mimic travel activities throughout the day. Second, LBSN data includes a sampling bias for different population groups and venue types. Third, the stochastic nature of human activities, especially the POI arriving patterns are critical for travel demand estimation. The existing approaches to the LBSN-based travel demand analysis have suffered from those limitations on deriving the dynamic travel demand patterns.
Recent development in spatial-temporal characteristics provides the opportunities to identify and quantify the correlation between LSBN-based travel activity and urban travel demand pattern. In this dissertation, a novel set of travel demand models based on the LBSN data is proposed and tested. The research starts with a comprehensive review of the existing travel demand data collection methods and the travel demand modeling. Then we introduce a profiling method to infer the functionality of city zones based on the POI categorical distribution and local mobility patterns. By classifying zones by these zone topics, we can now analyze interactions between zones of different functionality. Thirdly, by conducting zonal time-of-day variation modeling on the LBSN check-in arrivals, a new stochastic point process based trip arrivals estimation is developed. The output is applied to the input of a temporal delay based trip distribution model for deriving dynamic OD patterns. And the model calibration and applications are also provided and discussed. The evaluation results illustrate the promising benefits of applying LBSN Data in urban travel demand modeling.