Analysis and Modeling of Real World Car-Following Driving in Downtown Area Based on Variable Structured Piecewise Linear Model*

This paper proposes a new methodology for analyzing and modeling car-following driving behavior seen in downtown area based on real world driving data. The methodology leverages the hypothesis that at different distances to a leading car, the variables that influences car-following driving changes. The resulting model is a Variable Structured PieceWise Linear (VS-PWL) model where each piecewise region is defined based on range-band and the dynamics in each region is specified by linear regression model with different combination of input variables. The resulting car-following driver model has shown different choice of variables for each region. Finally, the proposed model is evaluated by comparing with Gipps model on the basis of real driving data prediction performance. Evaluation results show that the proposed model outperforms Gipps car-following driver model by 20%.