A Gaussian Kernel-Based Approach for Modeling Vehicle Headway Distributions

Abstract

Headway distribution models are essential for studying traffic flow theory, roadway accidents, and microscopic traffic simulations. Previous work has focused on parametric models. Vehicle headways were considered to follow some known parametric distributions based on certain assumptions. However, these assumptions are not universally acceptable and, consequently, the reliability of those headway distribution models varies significantly when applied to different flow conditions. In this study, a nonparametric distribution model with Gaussian kernel functions is introduced and assessed for vehicle headways on urban multilane freeways. Without any assumptions, Gaussian kernel models can extract intrinsic patterns from observed headway data to describe the distributing attributes of headways. Experiments were conducted to evaluate the accuracy of Gaussian kernel models for modeling vehicle headways. Results from the experiments indicated that the proposed models outperformed traditional parametric methods in a wide range of flow rates. Furthermore, transferability tests of the nonparametric model were performed, and the results showed that the proposed models can be generalized for applications at other locations with similar traffic flow patterns.