Methodology for Identifying Representative Data-Driven Urban Motorway Traffic Scenarios

Detection of traffic patterns from collected data plays a key role in identifying representative traffic scenarios. These scenarios are crucial when applying Machine Learning (ML) to traffic control systems, as they help ensure adequate controller performance under varying traffic conditions. This study addresses the challenge of selecting relevant scenarios for training ML-based traffic controllers by using data clustering methods for urban mobility analysis. To categorize and examine different traffic situations, unsupervised learning techniques, k-means and Self-Organizing Maps (SOM), were employed on real-world traffic data obtained from Slovenian motorways. The methodology for identifying relevant traffic scenarios was evaluated by applying reinforcement learning based variable speed limit control using the identified traffic scenarios as a training set. The effectiveness of the proposed data clustering methodology was tested on the unused traffic dataset. The findings provide insights for future studies involving larger datasets, incorporating data from multiple traffic detectors to develop relevant traffic scenarios for training ML-based traffic control systems.