The Recognition And Evaluation Of Freeway Traffic Safety States Based On Feature Selection

With our country’s transportation modernization and economic society coordinated developing in the 21st century,the scale of modern transportation infrastructure such as freeways has continued to grow,with the biggest mileage of more than 160,000 kilometers in the world.In addition,the car ownership ratio of urban and rural residents has also increased significantly.In 2020,the number of household cars per 100 households has reached 37.1.While the freeway network is improving day by day,the traffic volume is also increasing steadily.Among them,the most important issue is traffic safety.Studies have shown that freeway traffic accidents are highly correlated with the traffic safety states before the accident.Therefore,studying the freeway traffic safety states is of great significance to accident prevention and safety management.The paper establishes a feature selection,recognition and evaluation system of freeway traffic safety states based on features of traffic safety states.The main research contents include:(1)Obtain the sample data required for the research case with reference to the traffic data collection technology and standards,including freeway traffic flow data,traffic accident data,and meteorological data,then conduct quality analysis and evaluation of the data,and case-control study was used for data matching.(2)Study and improved laplacian score and random forest algorithm and then further designed integrated feature selection algorithm based integrated learning to obtain the individual feature ranking of specific road sections,through which the feature subset with the strongest relevance to itself is selected.Research shows that the integrated feature selection algorithm has higher stability and classification accuracy.(3)Based on the integrated feature selection algorithm,the paper designs an incremental learning algorithm of IFCM to recognize freeway traffic safety states.Bayesian conditional logistic regression method is used to evaluate risk with Markov Monte Carlo method used for sampling.Finally,the risk level of each traffic safety state is obtained.(4)The case of this paper is a section of Interstate 10 in the United States.The paper selects 5 features,recognizes 7 traffic safety states,and obtains the corresponding safety risk levels ranking,which can improve the accident warning ability of the freeway intelligent safety system,enhance the ability of managers to recognize safety risks,and take timely measures to change the current traffic safety state to improve freeway safety level.This dissertation includes 27 figures,34 tables and 81 references.