Research On Safety Evaluation Method Of Driving Behavior Based On Vehicle Network Data

In recent years,with the progress of social economy and people’s pursuit of high quality life,the vehicle ownership rate in China has increased significantly,which brings new challenges to road safety management.How to strengthen traffic safety management and prevent accidents from the driver’s behavior mechanism has become the focus of traffic research in recent years.In this paper,based on the natural driving data,the method of rule definition was used,identification bad driving behavior of drivers in driving process,and based on the identification results,the multi-model integrated machine learning method was used,realize the driving style classification and personalized driving behavior safety evaluation,in order to solve the problem of road safety,provide effective data analysis methods.The main contents of this paper are as follows:(1)A rule-based identification method of bad driving behavior was studied.Firstly,the natural driving data are preprocessed.According to the definition of dangerous driving behaviors,the bad driving behaviors are extracted and quantified.Then,the parameters of driving behavior characteristics are standardized by replacing absolute units with relative units to realize the reconstruction of driving characteristics,and finally the driving behavior relative characteristic parameters of all vehicles are obtained.(2)A classification model of driving style based on ensemble learning was built.In order to improve the instability and lack of accuracy of traditional models in the classification of driving styles,this paper transforms the clustering problem into a classification problem,and comprehensively utilizes the theory of multi-model fusion to realize the classification and prediction of driving styles.Firstly,the integrated clustering method based on Fuzzy C-means(FCM)and Spectral Clustering(SC)is adopted to cluster the driving behavior characteristic data,and the data samples are divided into labeled samples and unclassified samples.Based on the results of pre-classification,MVESKC,a majority voting integrated learning Classification method based on Classification and Regression Tree(CART),Support Vector Machine(SVM)and KNearest Neighbor(KNN),is studied.A variety of weak learners are used for learning and the majority voting integrated strategy method is adopted to realize the Classification model of vehicle driving style.(3)A driving behavior safety scoring model was built.The driving behavior of the driver is divided into multiple scoring units.The weights of each scoring unit are given by Analytic Hierarchy Process(AHP)and Entropy Weight(EW)respectively.Then the weights of AHP and EW are combined by ordinary least squares.Each scoring unit is mapped to a continuous interval of 0-100 points by the scoring function.Finally,the driving operation score of the driver is calculated according to the weight value of the scoring unit and the weighted calculation of the specific score of the scoring unit.After the above exploration and analysis of natural driving data,this paper forms a systematic analysis framework for Internet of Vehicles data,and verifies the effectiveness and advancement of the method used through experimental comparison and analysis.