Analysis And Evaluation Of Vehicle Driving Behavior Based On Private Car Trajectory Fusion

With the substantial increase in the income of urban residents and the changes in demand for private cars,the number of cars has increased substantially,which has brought pressure to the inherent traffic environment.Among them,the demand for private cars is highly consistent with the travel mode people usually choose,and private cars have gradually become the main body of urban transportation.However,the driving behavior characteristics of private car drivers usually show diversification,individualization,and subjectivity,which derive dangerous driving habits and become the main factor leading to most traffic accidents.Therefore,in order to build a harmonious urban traffic driving environment,it is of great research value and significance to analyze and evaluate the driving behavior of private car drivers.Based on the trajectory data of private cars,combined with the on-board diagnostic system,this paper uses the deep learning neural network method to mine and identify the data that can reflect the driving behavior of the driver in the trajectory data,and proposes a comprehensive evaluation based on the combination of fuzzy logic and Bayesian optimization methods.model.The main contents of this paper are as follows:Aiming at the problem of incomplete and inaccurate driving behavior data provided by vehicle trajectory data from a single data source,this article integrates vehicle status information such as acceleration and heading information provided by the on-board diagnostic system to provide a complete and comprehensive private car Trajectory fusion data.In order to accurately describe the various non-linear characteristics in the fusion data of private car trajectories,and to capture the differences between individual private cars.This research proposes a driving behavior analysis model based on multiscale convolutional neural network(Convolutional Neural Networks,CNN)to accurately identify and predict driving behavior in trajectory fusion data.Firstly,a competitive learning neural network based on Gaussian kernel density is used to recognize driving patterns.Then combined with multi-scale CNN to extract a variety of features in the trajectory fusion data,the resulting base network can make full use of the extracted multi-scale features to obtain more accurate driving behavior recognition results.Experimental results show that the proposed method performs better than the commonly used methods such as KNN,BP neural network,CNN and Na(?)ve Bayeian methods in driving behavior recognition based on trajectory data fusion.Aiming at the problem that the calculation of the risk degree of driving behavior is difficult to determine due to the obvious differences between individual private car owners,a comprehensive uncertainty evaluation model based on fuzzy logic is proposed.The model first uses the fuzzy set of custom driving behavior characteristics to determine the membership function that meets the risk level of driving behavior;then establishes the corresponding fuzzy rules to describe the driver’s driving behavior scores at different times and locations from multiple levels;Finally,the parameters of the simple feedforward neural network are optimized through Bayesian optimization,so as to obtain the risk assessment result of the driving of the private car owner.The comparative experiment is compared with the benchmark evaluated by experts,and the result shows that the performance of the proposed model is better than the current mainstream method in evaluating the risk level of the owner’s driving behavior on the fusion data of real private car trajectory.