Analysis Of Lane-changing Influencing Factors And Risk Prediction Research Based On Trajectory Data

Lane-changing is a fundamental driving behavior that determines the vehicles speed and changes in traffic flow,allowing drivers to change lanes to achieve the desired speed or driving comfort.However,lane-changing behavior is complex and is an interactive process with surrounding vehicles,and faces significant traffic safety problems.The high rate of motor vehicle fatalities caused by lane-changing has drawn attention from society,making the study of lane-changing behavior an important topic in the field of traffic safety.However,current research on lane-changing behavior lacks to construct a comprehensive system of influencing factors,and to predict the degree of risk of lane-changing.Therefore,it is necessary to apply statistical methods to extract each possible influencing factor from natural driving data,apply mathematical methods to calculate the degree of influence of each factor,and use machine learning methods to predict lane change decisions and the risk of lane change,so as to provide a theoretical basis for traffic safety regulation and improvement strategy formulation.Firstly,this paper analyses the data characteristics of the naturalistic driving dataset NGSIM,selecting 34 factors in terms of macro external factors,driver motivation,lanechanging vehicle characteristics and vehicle group influence,and classifying the data according to lane-changing motivation lane-changing location.Secondly,in order to analyze the role of the influencing factors and to take into account the heterogeneous effect of the data,a random parameter Logit model was used to model the severity of the risk of lanechanging,quantify the influencing factors and analyze the marginal effects.In addition,to further exploit the influencing factors and to make the study of lane-changing risk forward-looking,a prediction model based on LSTM neural network is proposed,dividing the model into a lane-changing decision prediction module and a lane-changing risk prediction module.The lane-changing decision prediction module introduces an attention mechanism to train the sample data and to study the driver’s lane-changing decision and the risk level of lane changes.Finally,this paper constructs a comprehensive system of lane-changing risk influencing factors,and finds that the mean heterogeneity of the gap between vehicles and speed and the variance heterogeneity with lane-changing steering angle,the lane-changing prediction model can better predict lane change and predict lane-changing risk.On the one hand,the research results provide new support to improve the research on the risk of lanechanging behavior,and on the other hand,the neural network prediction can be applied to the actual road safety analysis,which can help regulate lane-changing behavior and improve road management and operational safety.