Data-driven Safety Analysis Of Driving Behavior Of New Energy Vehicles

The technical method of driving behavior analysis usually refers to the calculation of a score or index of each driving behavior component(speed,acceleration,route,pedal pressing,etc.)in a weighted way based on different research purposes,to measure the user’s driving behavior,or to label the user’s driving behavior.Bad driving behavior will increase the risk of traffic accidents.Professional and reliable driving behavior safety analysis can not only help owners understand their driving behavior types and improve driving habits,but also be used in vehicle insurance and vehicle management.New energy vehicle is a new generation of vehicles that rely on pure electric power or use a hybrid oil electric drive mode.The driving data of new energy vehicles cover various elements of the vehicle’s driving process,such as instantaneous speed,component acceleration,usage of each pedal and turn signal,GPS position,mileage,and battery information.They have obvious multi-dimensional and temporal characteristics,and are typical multivariate time series data.However,in terms of various parameters in the driving process,it is quite different from traditional vehicles,which leads to that the current driving behavior analysis model cannot be well applied to new energy vehicles,and it is difficult for us to obtain appropriate expert knowledge for feature selection and data annotation of the original driving data of new energy vehicles.In summary,the safety analysis of driving behavior of new energy vehicles faces three problems: high data dimension,huge data volume,and difficult data labeling.However,the development of sensor technology and deep learning models has also brought new ideas such as large-scale data collection,full time recording schemes,and small sample learning of driving data to the analysis of driving behavior of new energy vehicles,providing opportunities for solving the prediction of driving behavior of new energy vehicles.In order to solve the three problems mentioned above with regard to driving data of new energy vehicles,and to conduct more professional and systematic research on the safety analysis of driving behaviors of new energy vehicles,this article,starting from the collected real driving data of new energy vehicles,studies the multiple time series modeling and feature importance of driving behaviors of new energy vehicles through several chapters,respectively,And the classification of driving behavior safety of new energy vehicles in small sample scenarios and self monitoring scenarios.Based on a large number of experiments,an efficient model has been improved and designed,and good experimental results have been achieved.On this basis,an intelligent transportation evaluation system including the classification of driving behavior safety of new energy vehicles is established.Specifically,the main contributions of this article are as follows:1)In order to make a more professional and systematic analysis of the driving behavior of new energy vehicles and solve the problems caused by their data dimension and quantity,this paper models them as a problem of multi-variable time series data classification,extracts the features of the data and labels them manually,and puts forward a deep learning model for the classification of driving safety of new energy vehicles.2)In order to solve the problem of data labeling and further improve the model perfor- mance,this paper studies the new energy vehicle driving behavior safety analysis in a few-shot scenario by data augmentation and transfer learning,and further im- proves the performance of the new energy vehicle driving behavior safety analysis model.3)In order to further adapt to the characteristics of high dimensionality and lack of labeling of driving behavior data of new energy vehicles,this paper further explores the multiple time series prediction problem under the scenario without additional labeling,and finally establishes a self-supervised pre-training framework based on Transformer.4)In order to verify the actual effectiveness of the driving behavior analysis framework for new energy vehicles,this article implements a user-friendly algorithm validation demonstration system,which implements a system pipeline on a unified platform to use and evaluate traffic prediction models.