Analysis of driving styles in cut-in behaviors on highway segments using trajectory data

Abstract

Vehicle Platoons (VPs) organize multiple vehicles to travel together, forming an efficient and coordinated moving group. For the foreseeable future, VPs and human-driven vehicles (HDVs) will coexist in mixed traffic flows, with frequent instances of VPs cutting into the trajectories of HDVs. To effectively understand and manage this cut-in behavior, it is crucial to grasp the underlying driving styles. Based on real-world natural vehicle trajectory data, this study explores the classification and analysis of driving styles during the cut-in process in vehicles, especially on highway segments. The research initially preprocesses a real-world natural dataset, including wavelet denoising. Fourteen features are selected from the speed-change preparation phase and lane-change phase, covering the state of the vehicle itself and its interaction with other vehicles. Next, Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) are used to reduce the dimensionality, followed by clustering driving styles using the K-means++ algorithm. The results show that for the selected features, t-SNE technique shows better clustering effect than PCA as a whole, with optimal clustering performance at four clusters. Therefore, we categorize the driving styles of highway segment vehicle cut-in into four types. Finally, the characteristics of cut-in process of different driving styles are analyzed statistically, which shows that it is effective to classify driving styles based on real natural data set.