Analyzing Microscopic Behavioral between Two Phases of Follower and Leader in Traffic Oscillation with Developing Artificial Neural Networks

Mirbaha, Babak; Kordani, Ali Abdi; Salehikalam, Arsalan; Akbarinia, Farzad; Zarei, Mohammad · 2017 · Crossref

DOI: 10.28991/cej-2017-00000110

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the microscopic behavioral dynamics of follower vehicles during traffic oscillations, specifically focusing on the transition between deceleration and congestion phases. Motivated by the negative impacts of stop-and-go traffic, such as travel delays and safety risks, the research aims to identify how driver behavior influences the propagation of deceleration waves. The authors address the limitations of existing theories, noting that while Newell’s car-following model provides a baseline for wave propagation, it fails to account for the nonlinear, asymmetric behaviors of drivers during acceleration and deceleration. Consequently, the study integrates asymmetric microscopic driving behavior theory for deceleration phases and traffic hysteresis theory for acceleration phases to classify driver behaviors into specific patterns, such as "over reaction," "under reaction," "aggressive," and "timid." The methodology utilizes vehicle trajectory data from the Next Generation Simulation (NGSIM) program, collected from Interstate 80 and US Highway 101 during periods of transient to congested traffic. Raw data were smoothed using the Savitzky-Golay filter. The researchers identified critical points of wave propagation and reception based on Newell’s theory and classified driver behaviors by analyzing deviations from ideal trajectories. To handle the nonlinear nature of driver responses, the study developed Multi-Layer Perceptron (MLP) artificial neural networks. These networks were trained to predict the time interval between the reception of a deceleration wave and the onset of congestion, using eight microscopic parameters as inputs, including spacing and speed differences at wave propagation and reception points. Sensitivity analysis was conducted using Crystal Ball software linked to MATLAB to determine the influence of these parameters on the time interval. The results indicate that the developed neural networks effectively modeled the relationship between driver behavior and phase transition times, achieving high correlation coefficients (0.93 for under-reaction/timid and 0.92 for over-reaction/timid patterns). For the "over reaction–timid" pattern, the time between deceleration and congestion phases ranged from 5 to 8 seconds. Sensitivity analysis revealed that the spacing difference between the deceleration and congestion phases was the most effective parameter influencing this time interval. Specifically, increasing the spacing difference resulted in a decrease in the time interval for under-reaction–timid drivers and an increase for over-reaction–timid drivers. Other significant parameters included the follower’s spacing at the wave reception point and the speed difference between phases. The significance of this research lies in its detailed characterization of microscopic driver behaviors during traffic oscillations, providing a more nuanced understanding than traditional car-following models. By identifying specific behavioral patterns and their corresponding effects on traffic wave propagation, the study offers insights into the mechanisms of congestion formation. The findings suggest that driver behavior, particularly regarding spacing adjustments during deceleration, plays a critical role in the timing and severity of traffic oscillations. This understanding can inform the development of more accurate traffic simulation models and intelligent transportation systems capable of predicting and mitigating stop-and-go traffic conditions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• following distance