A Deep Learning Model for Off-Ramp Hourly Traffic Volume Estimation. Issue 7 (July 2021)

• Record Type: Journal Article
• Authors: Nohekhan, Amir; Zahedian, Sara; Haghani, Ali
• Is Part Of: Transportation research record. Volume 2675:Issue 7(2021)
• Page Start: 350
• Page End: 362
• Publication Date: 2021-07
• Subjects: Transportation -- Periodicals; Roads; Transport -- Périodiques; Routes -- Périodiques; Routes -- Conception et construction -- Périodiques; Roads; Transportation; 388.05
• DOI: 10.1177/03611981211027151
• Languages: English
• ISSNs: 0361-1981

Abstract

This paper addresses estimation of traffic volume of freeway off-ramps. Freeways are the transportation network's main corridors, serving a large portion of the traffic volume. This traffic passes into the lower-level roads through off-ramps. Therefore, the traffic condition of the off-ramps is an essential factor affecting the operation of the transportation network. The continuous collection of volume data is impractical, and transportation authorities install vehicle detectors permanently on only a few off-ramps and temporarily (e.g., a week) on some others. Thus, traffic volume is the most challenging to estimate among various traffic measures. Moreover, the existing literature on volume estimation is mainly concerned with evaluating traffic on the main road segments. However, the distinct characteristics of the connection links, such as off-ramps, demands specified modeling. This study estimates the hourly traffic volume of off-ramps using a deep learning model. It evaluates the advantages of inputting the connected lower-level road features to the model, and explores various detector installation strategies on the model training process. The primary data sources are volume counts, probe speeds, and road segment infrastructure characteristics. The model results indicate that the incorporation of traffic flow characteristics and infrastructure attributes of the lower-level road connected to the freeway significantly improves the accuracy of estimation off-ramp traffic volume. Further, analysis illustrated that the model trained with data from temporarily installed detectors on all interchanges outperformed models trained with permanently installed detectors on 90% of the interchanges, indicating the model's ability in extracting temporal correlations significantly more than spatial correlations.