Risk assessment method for lane-changing vehicles based on surrogate safety measure

Abstract

Pre-accident risk assessment using surrogate safety measure (SSM) is an effective way to identify accident risk levels and promote accident management. Lane change is a typical driving behaviour. The management department can remind the driver to adjust driving behavior or revoke the intention of lane change when there is a higher risk in lane change, which can effectively reduce the accident rate of lane change and improve the road capacity. Therefore, to improve the accuracy of vehicle risk identification during lane-changing, this paper proposed a driving risk assessment method of lane-changing vehicle based on SSM. Firstly, based on the driving characteristics of lane-changing vehicles, such as velocity and acceleration, this study further built the risk feature, including Time to Collision (TTC), Time Integrated Time to Collision (TIT), Deceleration Rate to Avoidance (DRAC) and Crash Potential Index (CPI). Then the correlation between each feature was analyzed. Secondly, the Kmeans algorithm was used to classify the vehicle risk. According to the data distribution of the key driving features of the vehicle and the result of the cluster analysis, the risk level of the vehicle was evaluated during lane changing. Finally, this study used the eXtreme Gradient Boosting (XGBoost) to identify the risk of driving behaviours during lane changes. The performance of XGBoost under different optimization methods and other machine learning algorithms was compared. It was found that XGBoost risk identification accuracy rate based on Bayesian optimization algorithm was 95.65%, which could realize the accurate assessment of driving risk.