Abstract : The modern road network infrastructure (signal controllers and detectors) continuously generates data that can be transformed and used to evaluate the performance of signalized intersections. In order to automatically make meaningful observations about signal performance, we propose the application of data summarization and compression techniques in order to intelligently group together intersections and/or time intervals during the day and certain days of the week. This work details the use of linear and nonlinear dimensionality reduction techniques to achieve the aforementioned goals. The approach is also extended to perform change detection so that significant changes at intersections and corridors can be highlighted.
This work presents a novel framework that combines processing of high resolution controller log data pertaining to certain performance measures with modern data mining and machine learning techniques to produce the following outcomes:
- Automatically learn a compact representation of signal performance and behavior both in terms of the signal's capacity to serve demand and coordination of signal with upstream and downstream intersection
- The compact representation enables the grouping of signals based on similar demand patterns and performance over time and space as well as the ordering of these groups in terms of observed performance.
- Computing the evolution of these performance-based groups over space and time can help in deriving potential time periods for coordinated signal timing plans
- Our models can be used to discover significant changes in signal performance and hence to estimate the need to update signal timing plans. In other words, detect temporal changes in signal performance over multiple weeks or months and detect periods with changes in many signals