Inferring Daily Itinerary for Drivers Based on Spatial and Temporal Analysis

Daily itinerary, consisting of an individual's trips and activities on a day, is usually fundamental input for many travel demand models. However, current research lacks effective methods to extract complete daily itineraries of large-scale samples for a long period. To bridge this critical gap, this study presents a methodology to infer drivers' complete itinerary with Automatic Vehicle Identification (AVI) data. Itinerary inference issue is challengeable, as there are a large number of possible combinations of activities and trips and only a limited amount of observed data at the individual level. A problem-specific Graphical Network Model named ST-II is constructed by incorporating spatial and temporal analysis, which can implement an end-to-end processing and inference pipeline from the raw AVI data to the complete itinerary information. To seek the best matching itinerary with observed AVI data among vast feasible states, a candidate mobility state generation algorithm and optimal itinerary searching algorithm are developed. Empirical studies have been conducted based on field-test data. Compared with two benchmarks, the proposed ST-II improved the inference accuracy significantly even for activities with missing observations. Sensitivity analyses on the size of traffic area zone and the spatial gap between AVI observations have also been performed to provide guidance for administrations and researchers on the partition of the study region and placement of AVI sensors. This is the first research that attempts to infer the complete daily itineraries of drivers from AVI data. It can enable a series of applications and three examples are also presented in this paper