Driver Behavior Modeling And Characteristic Analysis Based On Trajectory Data Mining

As intelligent connected vehicle has become an important construction tendency in the national strategy of buidling a country with strong transportation network in China,urban intelligent transportation system infrastructure,such as intelligent vehicles,self-driving,and cooperative vehicle infrastructure system,plays a more and more important role in ensuring driving safety and improving travel efficiency.Intelligent connected vehicle is the core component of "human-vehicle-road-environment" travel ecology and is the tool of importance for drivers to participate travel acitivities,in consequence how to design an intelligent assitstant driving mode,which is more in line with the driver's behavior preference according to the driver's driving characteristics,is the key to realize the intelligent and shared travel mode of the city in the future.Therefore,in order to provide sufficient theoretical reference for driver-oriented intelligent network vehicle design,this paper takes driver behavior as the research object and builds a driving behavior model based on rich driving behavior data.This paper also explores the linkage relationship between driving style,driving environment factors,and driver behavior decision-making,and reveals how driver behavior characteristics are affected by individual driver factors and environmental interference factors.Firstly,the driving behavior features were extracted from the original vehicle trajectory GPS data by using feature engineering technology and mathematical statistics.The correlation between driving behavior,individual driver factors,and driving environment factors was discovered by analyzing the changing rule of behavior feature series data,so as to provide basic reference for driving behavior modeling and driving behavior characteristic analysis.Then,referring to the classical Markov process and Dirichlet process theory in the field of natural language processing,a hierarchical Dirichlet process-hidden semi-Markov(HDP-HSMM)driver behavior model was established based on nonparametric Bayesian learning theory.The HDP-HSMM model is able to transform the driving behavior feature series data into a behavior state series with the behavior semantics.Based on the sharing mechanism of behavior state space for multiple drivers,the distribution of driver behavior state series can reflect a driver's behavior preference.The modeling results can serve as a frame of reference for the analysis of driver behavior characteristics considering driver individual factors and driving environment factors.Finally,feature selection and feature clustering technologies were introduced and three driving styles of driver groups were identified by analyzing the driving behavior feature information.Multiple statistics approaches were applied to implement the feature variation analysis and the behavior pattern of driver groups was revealed from three aspects of driving feature including aggression levels,stability levels,and preference levels,and then the semantic comprehension of each driving style was ultimately given.Based on the established HDP-HSMM model,this paper examines the relationship between driving style factors,driving environment factors(including the traffic status,the traffic generation and attraction)and driver behavior,respectively.In the meantime,by mapping different driving environment levels into various driving scenes,this paper explores the performance of driver behavior under the comprehensive influence of driving style and driving environment factors.It provides an important theoretical support for illustrating the relationship between individual driver factors,driving environmental interference factors and driving behavior characteristics.There are 62 figures,9 tables,and 65 references in this paper.