Research On Driving Style Identification Strategy Considering Driving Conditions

Advanced Driving Assistance System(ADAS)can help drivers to reduce driving load,prevent traffic accidents,and get widespread application on intelligent vehicle.For ADAS,the driver’s experience is the top priority.Different drivers have different driving styles due to their gender,age,occupation,personality and other factors,and their experience in ADAS is also different.In order to improve ADAS’s comfort and safety,the research of driving styles has become a hot spot in recent years.In the current research,most researchers are mainly focused on driving styles and identification of driving styles under specific conditions.After getting the driving style label,the personalized ADAS function is implemented under specific conditions according to the characteristics of each type of driver.The realization of multi-operating conditions and personalized driving assistance still needs to conduct in-depth research on the consistency of driving styles between different working conditions and the choice of driving styles identification methods.Based on projects such as the National Natural Science Foundation of China(No.52172386,No.51775235),this paper has conducted research on driving styles identification strategies considering driving conditions.Based on real vehicles,a driving data collection platform was built.Driver data is collected under different driving conditions;A driving condition identification rule is proposed based on the yaw rate average energy and dynamic time warping algorithm;Factor analysis method is used to extract the characteristic parameters of driving behavior under multiple driving conditions;The K-means algorithm optimized by genetic algorithm is used to complete the clustering representation of drivers’ driving styles and explore the consistency of driving styles under multiple driving conditions;XGBoost algorithm based on Sparrow algorithm to complete the training of driving style identification model under multiple driving conditions;Finally,the driving style identification strategy considering driving conditions is obtained and tested using real vehicle data.The research content of this paper mainly includes the following four parts:(1)Driving data collectionBased on hardware such as RT3002&RT Range,d SPACE Micro Auto Box 1401,RTMaps,millimeter wave radar,and cameras,a real vehicle acquisition platform for driving data is built.Three typical driving conditions,such as car following,lane changing,and overtaking,are designed and social drivers are recruited for driving data collection.Use the driver simulator to complete the data acquisition of U-turn and turning conditions.Finally,the driving data is preprocessed to provide a data foundation for future research.(2)Driving condition identificationIn order to accurately identify three types of typical driving conditions from driving data,a baseline sequence library of five basic driving conditions is established:following,changing lanes,overtaking,turning,and turning around.Turn around conditions are eliminated through vehicle speed thresholds.Combining the average energy of yaw rate and the following distance to complete the identification of following conditions.Based on the dynamic time warping algorithm,lane changing,overtaking,and turning conditions are identified,and lower bound functions and early termination ideas are introduced to reduce the complexity of the algorithm.Finally,the collected condition data are used to verify the condition identification rules.(3)Clustering of drivers’ driving stylesSelect driving characteristic parameters based on the characteristics of different driving conditions,and use factor analysis to reduce the dimension of the characteristic parameters to obtain driving factors for different driving conditions.K-means algorithm based on genetic algorithm optimization completes driver driving styles clustering,and obtains driver driving styles labels under different driving conditions.The correlation analysis of driving styles under different driving conditions proves that there is no inherent correlation between the driving styles clustering results of drivers under multiple driving conditions,indicating the direction for training the driving styles identification model under multiple driving conditions.(4)Driving stylet identification strategy considering driving conditionsTaking into account the characteristics of different driving conditions,XGBoost algorithm optimized based on Sparrow algorithm is used to train driving style identification models for three driving conditions.Combining the driving style identification model for multiple driving conditions,driving condition identification rules,and driving feature factor extraction methods,a driving style identification strategy considering driving conditions is obtained.The accuracy of the driving style identification strategy considering driving conditions proposed in this paper is verified by the data collected by the acquisition platform.