Citation

Khajeh Hosseini M, Talebpour A. J. Transp. Eng. A: Systems 2023; 149(4): e04023019.

Copyright

(Copyright © 2023, American Society of Civil Engineers)

DOI

10.1061/JTEPBS.TEENG-7209

PMID

unavailable

Abstract

Traffic management strategies have been relying on various congestion prediction methodologies. The prediction accuracy of these methodologies has improved over the years, offering reasonable short-term and midterm predictions of macroscopic traffic measures (i.e., flow, speed, and occupancy/density). Unfortunately, by relying on fixed infrastructure sensors and aggregated data, these prediction methodologies fail to include microscopic traffic flow dynamics in their prediction algorithms. Accordingly, they usually fail to capture the onset of congestion and can only predict the propagation of existing shockwaves. That is, in fact, critical for utilizing effective traffic management strategies because predicting the onset of congestion can significantly help with mitigating it. Addressing this shortcoming in traffic predcition algorithms, this study proposes a deep learning methodology to predict the formation and propagation of traffic shockwaves at the vehicle trajectory level. Assuming the existence of communications between vehicles and infrastructure, the time-space diagram of the study segment serves as the input of the deep neural network, and the output of the network is the predicted propagation of shockwaves on that segment. It is the capability to extract the features embedded in a time-space diagram that allows this methodology to predict the propagation of traffic shockwaves. The proposed approach was tested on both simulation and real-world data, and results show that it can accurately predict shockwave formation and propagation.

Language: en