Driving Behavior Profiling Based On Private Car Trajectory Data

In the information age,the analysis of personal driving behavior,travel trajectory,long-term law and other vehicle related data information has become one of the most important research objectives of the current automobile traffic planning and development.The rapid development of China’s automobile industry,the emergence of new energy vehicles and the rapid popularization of on-board intelligent terminal equipment provide conditions for recording a large amount of track data,which also provides a data basis for large-scale research on driving behavior.Urban environments of drivers usually show the driving behavior characteristics of diversification,personalization,thus constitutes a long-term habitual driving behavior,study and analysis of drivers’ long-term driving behavior characteristics is of great significance for t he detailed management of urban road traffic safety,which helps drivers to have a more macroscopic understanding of their long-term driving habits,and further evaluation of drivers’ driving risks also provides a reference for the development of personalized auto insurance premium programs..In this paper,based on the analysis and research information from the vehicle trajectory data,through the GPS devices and on-board diagnosis system,combined with risk driving algorithm of trajectory data,which can identify individual driving characteristics of drivers with deep learning method for mining and recognition,according to the driving behavior characteristic index and other trajectory data of driving style orientation identification,Construct neural network classification model.Finally,a driving behavior score prediction model based on historical driving evaluation data is proposed.The main research work is as follows:(1)In order to accurately judge all kinds of risky driving behaviors,a recognition algorithm of risky driving behaviors is proposed.It can accurately identify the behaviors of sharp acceleration and deceleration,sharp turns and overspeed in the process of vehicle driving,and set up a specific experiment to collect predefined trajectory data of test vehicles to verify the accuracy of the recognition algorithm.(2)In order to accurately capture the personality characteristics of private car drivers,this paper proposes a driving type recognition model based on BP neural network.Based on the driving behavior data,the vehicle data and risky driving behavior parameters needed to identify driving types were extracted,and the sample data were classified by k-means clustering algorithm,and three feature labels were extracted.Based on the sample results of clustering,the BP network model has more accurate identification performance,and the accuracy can reach 94.04%.The type classification and recognition of any driver track data can provide a basis for the driving behavior score predict ion in the following paper.(3)In order to quantitatively evaluate the driving risk of drivers,a driving behavior score prediction model was constructed,which was based on the input historical data of driving behavior score.The core of the prediction model is temporal graph convolution network(TGCN),which uses GCN to model the urban road network through the combination of GCN and GRU.The road spatial characteristics are captured by GCN and the time correlation of driving behavior score of input trajectory data is captured by GRU.The model captures the temporal and spatial features from the historical score data to predict the future driving score more accurately.Finally,HA model,SVR model and GRU model are introduced to compare the results.The results show that TGCN model has better performance in different time prediction ranges and is superior to the current mainstream prediction methods.It can provide some reference for the formulation and evaluation of UBI.