Characterization Of Driving Behaviors And Its Applications

With the development of traffic system and the increase of vehicle,road safety has become one of the most important problem for the society.The factors that affect the driving characteristics of the driver mainly include the driver's own factors,vehicle factors,environmental factors and so on.Therefore,there are many difficulties in the study of driving behavior.This research combines machine learning and statistics related methods,and proposes a series of methods for modeling and quantitative evaluation of driving behavior.The main content of the paper includes:(1)A normalization algorithm for driving behavior based on a gated sparse self-encoding network(GSAE)is proposed.This method learns the driving behavior sequence and the speed of the vehicle at the same time,and maps the driving behavior to the speed curve of the standard vehicle test.The corresponding normalized driving behavior is obtained during the learning process.The proposed method is simple and effective,and overcomes the shortcomings of traditional methods which takes time and effort to build models for the driver and the vehicle one by one.(2)The phase space reconstruction method is used to reconstruct the phase space of one-dimensional driving behavior,and then the trajectory image in the phase space is processed with the pre-trained convolutional neural network for feature extraction.Later on,the UMAP technique is used for high-dimensional vector for dimensionality reduction and visualization.The effectiveness of the proposed method in characterizing driving behavior is verified by the distance measurement and driving style recognition task.(3)Based on the semantic unit of shift behavior,the symbol aggregation approximation(SAX)method is used to symbolize the shift behavior.On the basis of reducing the dimensionality of the data and retaining the original characteristics,the distribution of the data is also retained.Therefore,it provides a new effective scheme for the semantic analysis and quantitative evaluation of driving behavior.The proposed method is successfully applied in the drivers' recognition task.(4)The differences between the driver's manual shift preference and the existing shift schedule baseline is analyzed;A comprehensive evaluation index for the shift schedule is established in consideration of fuel consumption and shifting quality;By using the bionic algorithm,the online evaluation and optimization are conducted,and the most suitable shift schedule in consideration of shift preference is obtained through iterations.