Prediction Of Dangerous Car Following State Based On Gaussian Mixture-Hidden Markov Model

With the continuous increase of the total number of private cars and the number of motor vehicle drivers in China,more and more problems such as road congestion and frequent traffic accidents have been brought about.As a common driving behavior,most of the traffic accidents are caused by vehicle following or related to it,so the research on the prediction of dangerous car-following state is of great significance.This study uses the Safe Pilot Model Deployment(SPMD)dataset to predict the dangerous car-following state based on the Gaussian mixturehidden Markov model and the random forest model through the observable information such as the motion parameters of the vehicle and the preceding vehicle.The results of this study can assist drivers to make early warnings and reduce the occurrence of dangerous car-following accidents.Firstly,this study uses the SPMD natural driving dataset for car-following data selection and processing.,and this research chooses the inverse time-to-collision(ITTC)threshold for measuring the dangerous car-following state of vehicles by reading the relevant literature on dangerous car-following states and collision risks at home and abroad and then the samples of dangerous car-following and safe car-following were picked out.Secondly,in order to reduce the error and make full use of the observation data,this research takes 0.5s as the time window length.Through the quantification of the car-following state in the time window and the eigenvalue processing of the motion parameters,this research uses the grey correlation method to select observation variables that are highly correlated with the car-following state for combination.During model training,in order to ensure the rationality of the initial parameters of the Gaussian mixture-hidden Markov model,on the basis of statistical analysis,this study used the Gaussian mixture model to fit the observed variable data to obtain the initial parameters of the model and train them,which guarantees the accuracy of the training results.Then,this study optimized the model for different combinations of observed variables,different numbers of Gaussian mixtures,and different characteristic parameters.After obtaining the optimal model,this study predicted the dangerous car-following state in advance of 0-5 time windows to compare the prediction results of 0.5-2.1s in advance,and according to the results this study selected the best advance prediction time in combination with practical applications.Finally,this research compared the prediction effect of the Gaussian mixture-Hidden Markov dangerous car-following state prediction model by constructing a random forest comparison model.The results proves that the Gaussian mixture-hidden Markov model has good applicability to the data with time series characteristics,and the Gaussian mixture-hidden Markov model can predict the dangerous car-following states well.