Event-driven Driving Style Classification And Application

With the advent of the era of big data,autonomous driving is gradually entering everyone’s daily life.As an important participant in traffic,the driver’s behavior affects the vehicle’s fuel economy,driving safety and the realization of related vehicle assistance functions.How to ensure the safety and comfort of autonomous vehicles is a common problem faced by researchers.It is not only necessary to ensure that the vehicle runs safely under the premise of obeying by traffic rules,but also to ensure the driver’s personalized experience during the driving process.A good personalized experience can improve driving comfort and ensure driving safety.The research on driving style can reflect the different characteristics of drivers to a certain extent,so as to improve the driver’s acceptance of various driving functions,thereby bringing a better personalized experience.In the study of driving style,current research performs feature calculation based on data,and then use rule-based methods,neural networks or deep learning methods to classify driving style.However,these methods ignore the characteristics of driving style,different external conditions and different driver’s own conditions will cause changes in driving style due to the fact that driving style is a characteristic of the driver.Driving style strongly constrains fuel consumption and safety,and is key to understanding drivability and ensuring acceptance of advanced driver assistance systems and automation.Therefore,considering the variety of driving style,this paper proposes an event-driven driving style classification method creatively,which can intuitively see the driver’s style tendency during the driving process.What the paper mainly researched is just as follows:（1）First,the relevant information of the driver research is combed in detail.Explain the differences and connections between driving behavior,driving style,driving skill,and driving pattern.A variety of data that can be used in driving behavior research are also presented.In order to better collect more personalized driving data,this topic collects CAN data through real vehicle experiments for subsequent analysis and research.（2）Carry out research on event-based driving style classification.The categories and definitions of driving events are briefly introduced from the lateral and vertical perspectives,respectively,and the turning events and ZTS events are extracted according to the definitions.The high-dimensional data visualization algorithm and kmeans clustering algorithm were used to determine the style center,and then the principal component analysis was used to determine the event center by comparing the contribution rate of the first principal component.According to the principle of the closest distance,the style factor of each driver was determined.Then,the horizontal and vertical style factors are comprehensively weighted by the analytic hierarchy process and the entropy weight method,and the overall style tendency of different drivers is obtained.（3）Driver identification based on driving style factor.The definition of the BP neural network and the basic structure of the neuron model are briefly introduced,and the process of using the formula to derive the network training is shown in detail.At the same time,the network training effect under different numbers of hidden layer nodes is compared,and the number of nodes with the best training effect is determined.Based on these,a neural network is constructed.The network input features are constructed based on EMD decomposition,and the first three Intrinsic Mode Functions are used to calculate the features,which can not only exclude the interference of noise,but also enrich the feature samples.The recognition result of the network are compared with and without considering the style factor,respectively.The result shows that the recognition result considering the driving style factor can reach 98%.To sum up,the event-based driving style classification method proposed in this paper takes into account the diverse characteristics of driving styles,so that driving styles can be evaluated more accurately and reliably.At the same time,the style factor obtained from the research can be used as a new dimension feature of driver identification,which can effectively improve the accuracy of driver identification.Therefore,in the era of rapid development of intelligent driving,the method proposed in this paper has good practicability in improving the personalization,safety and comfort of driving.