Abstract : he road network infrastructure (signal controllers and detectors) continuously generates data that can be transformed and used to evaluate the performance of signalized intersections. Current systems that focus on automatically converting the raw data into measures of effectiveness have proven extremely useful in alleviating intersection performance issues. However, these systems are not well suited for automatically generating recommendations or suggesting fixes as needed. In this work, we demonstrate the use of machine learning and data compression techniques to build a recommendation system. Specifically, we present an end to end solution for automatically generating intersection coordination plans.
Coordination involves synchronizing multiple intersections to enhance the operation of directional movements in a system.
Typically, the green times for a set of intersections are synchronized along the primary directions of movement using (speed-dependent) timing offsets to account for the travel time between intersections. Such coordination of intersections can result in improvements in the quality of traffic flow along an arterial or street. The current practice for coordinating and re-timing intersections remains a largely manual process and is based on temporally limited data samples. In this paper, we present a recommendation system that generates intersection coordination plans using high resolution intersection controller logs collected over large time periods.
Our focus in this paper is to design an automated solution for the identification of intersections on a corridor that are good candidates for coordination and the time periods they should be coordinated for. Our approach has two major steps:
- Building accurate, descriptive models for performance measures of interest using time series modeling for all intersections under consideration. This is done to ensure that conclusions are not based on temporally local observations because they may be susceptible to outliers.
- Use of these descriptive models (based on key MOEs) to cluster the intersections in a corridor with similar traffic patterns (or demand). This is done for all corridors under consideration. We use the results of clustering and spatial information within a corridor to automatically deduce contiguous regions in a corridor or sub-corridor where the intersections should be coordinated and the times of the day and days of week they should be coordinated for.