A Research On Driving Behavior Modeling Method Based On Driving Style Characteristics

In recent years,the technology of intelligent and connected vehicles has become a research hotspot,and virtual simulation testing plays an important role in the development of intelligent and connected vehicles.The authenticity of the virtual test scenario is still a technical problem,and this paper aims to improve the authenticity of the virtual test scenarios and diversity.Based on the NGSIM(Next Generation Simulation)real vehicle trajectory data,this paper analyzes the driving style,extracts the distribution characteristics of the driving style feature,establishes a motion preview model based on driving style characteristics,and applies the model to the traffic vehicles in virtual test scenarios.Eventually,the traffic vehicles’ driving behavior can reflect the different driving styles.The research contents of this paper mainly include:First,this paper reconstructs the NGSIM vehicle trajectory data to provide accurate data support for driving style analysis.NGSIM vehicle trajectory data is an ideal data source for developing traffic simulation models,but there are errors and noises in it,which will amplify the errors and noises when calculating the speed and acceleration.Therefore,this paper reconstructs NGSIM vehicle trajectory data based on the two-step trajectory reconstruction method proposed in the literature [1]:(1)Identifying and correcting outliers: firstly,we identify trajectory outliers with wavelet transform,and correct trajectory outliers with Gaussian kernel regression.Then,we use physical constraints to identify acceleration outliers,and use cubic spline interpolation to correct the corresponding trajectory;(2)Using the wavelet transform threshold denoising to remove the speed noise.Finally,we verify the results by analyzing velocity,acceleration,its spectrum,and acceleration change rate.The results show that we remove the outliers and noises effectively under the premise of keeping the origin characteristics of vehicle trajectory,speed,and acceleration unchanged.Secondly,this paper analyzes the driver’s driving style in straight driving conditions according to NGSIM data,and uses probability density function to express the characteristics of driving style.Aiming at the problem that the semantic description of the clustering results is not easy to be accurate and unified due to the combination of horizontal and vertical features clustering,a driving style clustering analysis method based on separate horizontal and vertical features was proposed.We select driving style characteristics and use principal component analysis to reduce dimensionality.Then,we consider horizontal and vertical driving styles separately.K-means clustering is used to the first cluster according to vertical features and then we cluster the drivers according to horizontal features.Eventually,we divide driving styles into four categories: vertical radical and horizontal radical,vertical radical and horizontal conservative,vertical conservative and horizontal conservative.Then,the driving style characteristics of each type of driver are statistically analyzed.The results show that the characteristics of speed,acceleration,acceleration rate of change and lane centerline deviation of drivers with different styles are different.Finally,we fit the probability density function of driver lateral and longitudinal acceleration and lane centerline deviation to express the driving style of different drivers.Thirdly,this paper establishes a motion preview model based on the characteristics of driving style.We model the driving style characteristics based on the motion preview model based on the direction and speed integrated decision.We establish the following indexes of lateral random deviation to make the model follow the lateral displacement deviation generated randomly according to the different probability distribution characteristics,and to make the different motion preview models reflect different lateral displacement.Then,we establish the evaluation index of acceleration tendency to make the model follow the acceleration tendency according to the probability distribution characteristics so that the motion preview models of different styles can reflect the difference of acceleration.Finally,this paper verifies the results by co-simulation using Matlab/Simulink and VTD.The results showed that the model can reflect the difference in driving style of the drivers in different categories and the random difference of lateral deviation in the same category.