Estimating switching times of Actuated Coordinated Traffic Signals: A deep learning approach
Acceleration and Deceleration at signalized intersections are a major hindrance to vehicle fuel efficient operations. Green Light optimal speed advisory (GLOSA) allows controlling vehicles in a fuel-efficient manner but requires reliable estimates of signal switching time. This study aims at utilizing data from actuated coordinated signalized intersections in North Virginia along with multiple deep learning and machine learning techniques to provide estimates of traffic signal switching times from green to red and vice versa. These estimates can be used to enable more fuel-efficient operation using GLOSA and eco-driving. They can also be used to mitigate dilemma zone safety concerns. A comparative analysis will be conducted between the different techniques used and their pros and cons in terms of prediction errors and robustness to different traffic conditions.
Universities and Sponsoring Organizations Involved
U.S. DOT Office of the Secretary/Research
Funding Sources and Amounts
USDOT: $60,000 (Federal), Virginia Tech: $30,000 (Match)
October 1, 2020
Expected Completion Date
Sept. 30, 2021
Expected Research Outcomes
Machine learning models and frameworks for their implementation on signalized intersection, comparative analysis of different models in the context of actuated signal switching time prediction
Expected Equity Impacts and Benefits of Implementation
These models can be enabling technologies for Green Light Optimal Speed Advisory leading to more sustainable and more fuel-efficient operations. Moreover, the predictions can be used to modify the signal phase and timing (SPaT) messages allowing for more efficient operations for connected vehicles when traversing signalized intersections.
Traffic flow theory, Traffic Signal Operations, Machine Learning, Deep Learning, Artificial Intelligence, Big Data.