Understanding Drivers' Travel Behaviors through Vehicle Onboard Diagnostic Data Using Multi-Dimensional Discrete Hidden Markov Model

摘要

Conventional travel behavior research relies on questionnaires to understand individual travel patterns. However, these studies consume considerable manpower and resources. In a connected vehicle environment, on-board diagnostic (OBD) devices can record engine status, trajectories, acceleration, and fuel consumption of a vehicle to capture long-term variability in driver behaviors. In this study, the heterogeneous travel patterns of drivers are modeled using 2-month OBD data. An algorithm called clustering by fast search and find of density peaks is employed to classify drivers into long-distance and occasional, high-frequency, and regular travelers. The average travel distance, travel days, and first and last departure time records are considered in the procedure. A multi-dimensional discrete hidden Markov model is used to predict the category of any driver based on their historical travel behavior. This study provides useful data sources for activity-based modeling and also demonstrates the potential of vehicle OBD data for developing targeted online services.