Categorizing Car-Following Behaviors: Wavelet-Based Time Series Clustering Approach

Abstract

The categorization analysis of car-following behaviors is beneficial to enrich the current car-following models and the applications of connected and automated vehicles (CAVs) in a mixed traffic environment. Previous studies categorized the car-following behaviors during the traffic oscillations using artificially designed behavior patterns, but they are not quietly flexible and are limited to distinguish the complicated car-following behaviors. To address such a problem, the study proposes a wavelet-based time series clustering approach to automatically categorize the car-following behaviors. First, the response time series of the car-following behaviors are extracted using general Newell’s car-following model. Second, the discrete wavelet transformation algorithm is employed to extract the following behavior features from the original time series. Finally, the hierarchical clustering algorithm is used to categorize the car-following behaviors according to the calculated similarity between the transformed time series. Numerical tests on Next Generation Simulation (NGSIM) show that the proposed algorithm can effectively and automatically categorize the typical car-following behavior patterns summarized in the previous studies. The proposed algorithm is also flexibly implemented to discover the potential car-following behavior patterns. Findings suggest that a wavelet-based time series clustering by combing the Haar wavelet transformation algorithm and hierarchical clustering algorithm is a superior approach to automatically categorize car-following behaviors with a time series trajectory.