Pattern recognition using clustering algorithm for scenario definition in traffic simulation-based decision support systems

This paper presents a scenario clustering approach intended to mine historical data warehouses to identify appropriate scenarios for simulation as a part of an evaluation of transportation projects or operational measures. As such, it provides a systematic and efficient approach to select and prepare effective input scenarios to a given traffic simulation model. The scenario clustering procedure has two main applications: travel time reliability analysis, and traffic estimation and prediction systems. The ability to systematically identify similarity and dissimilarity among weather scenarios can facilitate the selection of critical scenarios for reliability studies. It can also support real-time weather-responsive traffic management (WRTM) by quickly classifying a current or predicted weather condition into pre-defined categories and suggesting relevant WRTM strategies that can be tested via real-time traffic simulation before deployment. A detailed method for clustering weather time series data is presented and demonstrated using historical data. Two clustering algorithms with different similarity measures are compared. Clustering results using a K-means clustering algorithm with squared Euclidean distance are illustrated in an application to travel time reliability.