A Framework for Evaluating Driving Patterns of Automated Vehicle Based on On-Road Driving Records

In this study, we propose a methodology for driving pattern analysis based on a hybrid clustering algorithm that combines the self-organizing map (SOM) and the K-means++ algorithm. The methodology aims to identify automated vehicle (AV) driving behaviors, highlight road sections with risky behaviors, and evaluate these behaviors using real-world driving data from automated shuttle operations. By employing a hybrid clustering algorithm grounded in Artificial Neural Network (ANN) principles, the approach enhances reliability compared to traditional clustering techniques. The analysis identifies both normal and risky driving behaviors, delineating specific driving patterns and visualizing them on maps. This enables the development of adaptable models for various AV types. Applied in the Pangyo area, the methodology reveals six distinct clusters, uncovering three risky behaviors and six unique driving patterns. This practical approach supports monitoring AV behaviors and improving road safety in mixed-traffic environments with AVs and conventional vehicles. Moreover, the methodology is adaptable to any region with high-definition (HD) map data, offering valuable insights for automated driving traffic control centers.