Online Competitive Algorithms and Reinforcement Learning for Traffic Management
Traffic signals with fixed-time signal timing require retiming often because of traffic volume changes. Traffic signals with a fixed time timings are configured with historical traffic data observations. The traffic signal retiming projects can take several years due to the major undertaking by transportation agencies. This seldom signal retiming can lead to higher vehicle delays. Although vehicle detection is a common technology, many intersections in urban areas are not setup with detection sensors because of their added installation and upkeep costs. In the Las Vegas Valley, there are fixed time and actuated (sensor-activated) traffic signals that we analyze for traffic congestion and to identify suboptimal signals. We built geospatial maps to perform exploratory visual analysis and to generate a list of the most monthly recurrent congested intersections in 2025. We also built an Autoencoder to find anomalous signals that are most likely experiencing congestion and identify important features that influence that congestion. Lastly, we present preliminary results from a traffic signal simulation that reveals that smaller cycle times only worsen average vehicle wait times by a small percentage from the optimal cycle length. The impact of our research has the potential to improve delays for motorist and to aid transportation agencies with signal retiming efforts.
Infrastructure Assets: Highway Assets
Resource Types: Research Report
Capabilities: Data & Information Systems
Management Processes: Monitoring & Adjustment, Performance Based Planning & Programming