Analysis of Lane Change Characteristics and Risk Clustering of Expressway Merging Bottleneck based on Trajectory Data

Abstract

In order to get a better understanding about lane change (LC) maneuvers, a total of 620 lane change samples were extracted by utilizing the vehicle trajectory data obtained by the unmanned aerial vehicle (UAV). Then, the distribution characteristics LC duration, LC distance, relative velocity, relative distance, Time to Collision (TTC) and LC starting points frequency are obtained. The distributions of LC duration and LC distance follow to the lognormal distribution. The proportion of the relative distance between the target vehicle and the leading vehicle on the current lane in the interval of [0m, 20m] is 61.90%. In 63.67% of the cases, the target vehicle's velocity is greater than the leading vehicle on the current lane. It indicates that the target vehicle chooses to change lanes when the driver cannot achieve the desired speed and driving conditions on the current lane. Finally, the LC collision risk clustering between the target vehicle and the leading vehicle on the target lane is clustered by Fuzzy C-Means Algorithm based on Genetic Algorithm and Simulated Annealing Algorithm (GASA-FCM). LC crash risk is divided into three categories, namely low risk, medium risk and high risk. The clustering results are consistent with the actual situation, demonstrating the effectiveness of the GASA-FCM clustering algorithm.