United States Department of Transportation
i
Enabling GLOSA through Domain Knowledge Aware SPAT Prediction and Queue Length Aware Trajectory Optimization
-
2024-08-29
[PDF-1.72 MB]
Details:
-
Creators:
-
Corporate Creators:Virginia Tech Transportation Institute ; Virginia Polytechnic Institute and State University. Charles E. Via Department of Civil and Environmental Engineering ; Morgan State University. Sustainable Mobility and Accessibility Regional Transportation Equity Research Center (SMARTER) Region 3 University Transportation Center (UTC)
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final, September 2023 - August 2024
-
Corporate Publisher:
-
Abstract:This research presents a novel approach to enhancing Green Light Optimal Speed Advisory (GLOSA) systems by predicting signal phase and timing (SPaT) switching times and assessing the confidence in these predictions. The proposed architecture leverages transformer encoders, combined with various deep learning methods such as Multilayer Perceptrons (MLP), LongShort Term Memory (LSTM) networks, and Convolutional LSTM (CNN-LSTM) networks, to form an ensemble of predictors. This ensemble is utilized to predict SPaT information for six intersections along the Gallows Road corridor in Virginia, addressing three primary tasks: predicting phase changes within 20 seconds, determining the exact switching time, and assigning a confidence level to these predictions. The experiments demonstrate that the transformer-based architecture outperforms traditional deep learning methods, achieving 96% accuracy in phase change prediction and a mean absolute error (MAE) of 1.49 seconds in exact time prediction. The ensemble predictions, particularly those with high consensus, are highly accurate, being within one second of the true value 90.2% of the time. In parallel, the study explores the implementation of GLOSA in proximity to actuated traffic signals, focusing on optimizing vehicle trajectories by incorporating real-time queue estimation from loop-detector and probe vehicle data. Simulation experiments evaluate the fuel savings potential and system performance both at the individual vehicle level and across the network. The results reveal substantial fuel savings of up to 35.7% for individual vehicles when considering queueing, with a saving of 10% for the case of an isolated intersection. However, the integration of queue estimation, while beneficial for individual vehicles, does not lead to significant network-wide performance improvements. This research underscores both the advantages and limitations of GLOSA systems, highlighting the importance of integrating real-time traffic data to optimize vehicle trajectories and improve system performance in real-world traffic scenarios.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
