Hybrid Recurrent Neural Network Modeling for Traffic Delay Prediction at Signalized Intersections Along an Urban Arterial

Subramaniyan, Arun Bala; Wang, Chieh; Shao, Yunli; Wan, Li; Wang, Hong; Zhang, Guohui; Ma, Tianwei · 2022 · OpenAlex-citations

DOI: 10.1109/tits.2022.3201880

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the challenge of predicting traffic delay at signalized intersections to facilitate real-time traffic control. While existing machine learning models offer high prediction accuracy, their complex, non-linear architectures often result in high computational burdens, making them unsuitable for real-time signal optimization. The authors propose a Hybrid Recurrent Neural Network (HRNN) model that balances prediction accuracy with computational efficiency. This approach is designed to predict one-step-ahead traffic delay while maintaining a linear representation of the control variable (green light duration), thereby simplifying the optimization problem for real-time signal control implementation. The study focuses on 34 signalized intersections along the Ala Moana Boulevard and Nimitz Highway arterial in Honolulu, Hawaii. Data were collected using the Centracs Signal Performance Measures system, which utilizes video cameras and magnetic detectors to record high-resolution traffic events. The authors developed a data processing pipeline to extract structured inputs—current delay, traffic volume, and green light duration—from unstructured raw event data. This process included handling missing data discrepancies between advanced and pulse detectors by generating artificial detector calls. Traffic delay was calculated based on estimated vehicle queues derived from arrival and departure patterns. The HRNN model combines a linear component for the control variable with a non-linear Recurrent Neural Network component to capture the stochastic, spatio-temporal dynamics of traffic flow. The model was trained using data from March to May 2021 and validated on June 2021 data, focusing on evening peak hours. The results demonstrate that the proposed HRNN model outperforms several existing modeling approaches, including standard linear models, Hybrid Neural Networks with Multi-Layer Perceptrons (HNN-MLP), and other recurrent variants like Hybrid LSTM and Hybrid GRU. The HRNN achieved superior performance in terms of Mean Absolute Percentage Error (MAPE) for both training and testing datasets. Crucially, the hybrid structure allowed for efficient computation, addressing the runtime burden associated with fully non-linear deep learning models. By isolating the control variable in a linear term, the model enables the use of efficient linear quadratic optimization techniques for signal timing plans, which is difficult with purely non-linear neural networks. The significance of this work lies in its contribution to intelligent transportation systems by providing a modeling framework that is both accurate and computationally feasible for real-time applications. The HRNN approach effectively captures the non-linear stochastic nature of traffic while remaining compatible with real-time control algorithms. This facilitates the development of closed-loop signal control systems that can dynamically adjust signal timing to alleviate congestion and improve traffic flow efficiency on urban arterials. The study validates the practicality of hybrid modeling for large-scale traffic networks, offering a viable path toward implementing AI-based traffic control in real-world infrastructure.

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.

• telematics crash prediction