Quantifying risks of lane-changing behavior in highways with vehicle trajectory data under different driving environments

Improper lane-changing behavior may have a great impact on the safety of the surrounding vehicles, and accurate risk assessment of lane-changing behavior can detect improper lane-changing behavior. Many existing risk assessment methods are based on the traditional minimum safe distance model as the theoretical foundation, not taking into account the effects of different weather and road conditions. Moreover, the initially calculated safe lane-changing distance through the model is generally the limit safety distance, which deviates to some extent from the initial distance maintained when the vehicle begins to change lanes during the actual driving process. The ratio of these two distances can reflect the degree of risk between the changing vehicle and its neighboring vehicles during lane-changing. For this reason, a lane-changing safety coefficient is defined and lane-changing risk assessment model based on the hyperbolic tangent function is constructed by combining the change characteristics of this coefficient. An environmental adjustment factor was introduced into the model to consider the influence of different driving weather and road conditions on the lane-changing risk assessment. Then, based on survey data from the survey questionnaire, the weights of the influencing factors were determined by using the hierarchical analysis method, and the environmental adjustment factor in the model was calibrated by using the fuzzy comprehensive assessment method. Finally, to verify the effectiveness of the model, the lane-changing trajectories of 373 vehicles were extracted from the HighD dataset. The risk assessment value of lane-changing was calculated by using the constructed model and was classified into high-risk, medium-risk and low-risk by using the K-means++ clustering algorithm. Various evaluation results were compared with the evaluation results from the expert evaluation method and the classical risk evaluation indexes, and the evaluation results are basically the same, which verifies the effectiveness of the lane-changing risk model established.