Revisiting Classical Two-phase and Kerner Three-phase Traffic Flow Theories: A Comparison of Pre-crash and Normal Traffic Conditions

TL;DR

This study compares classical two-phase and Kerner's three-phase traffic flow theories under pre-crash and normal conditions, revealing differences in jam propagation and data fit, which can improve crash prediction models.

Contribution

It reevaluates classical traffic theories in pre-crash conditions using real data, highlighting variations in jam dynamics and model fit compared to normal traffic.

Findings

Nearest downstream detectors better fit pre-crash data

Pre-crash wide-moving jams propagate faster

Higher standard deviation in jam velocities pre-crash

Abstract

Extensive research has been conducted to develop statistical and artificial intelligence-based models for predicting short-term crash probabilities using fundamental traffic flow variables and their associated descriptive statistics and mathematical transformations. However, there has been limited exploration into whether and how the fundamental relationships within traffic flow theories vary between pre-crash and normal traffic conditions. This study reevaluates four classical two-phase traffic flow theories and employs two methods from Kerner's three-phase traffic flow theory to compare their characteristics in the context of pre-crash and normal traffic conditions. The investigation is centered around the Shibuya 3 and Shinjuku 4 routes within the Tokyo Metropolitan Expressway. Data from both crashes and detectors was collected over a six-month period, spanning from March 2014 to August 2014. The findings reveal that data from the nearest downstream detectors to the crash locations provided a superior fit for pre-crash data compared to the upstream detectors. Notably, a noticeable decrease in goodness-of-fit was observed when compared to normal traffic conditions. In pre-crash scenarios, wide-moving jams exhibited a faster propagation from downstream to upstream, distinct from the patterns observed in normal traffic conditions. Furthermore, pre-crash data displayed a higher standard deviation in the calculated wide-moving jam velocities compared to normal traffic conditions. These insights have the potential to be highly valuable in the development of predictive crash models and in the estimation of traffic volumes on freeways across various timeframes, accounting for both free-flow and congested traffic scenarios.