Research And Application Of The Accident Risk Prediction Model Of Mountain Expressway Based On Machine Learning

Mountain expressway accidents occur frequently,and the prediction accuracy of existing research results is low,and the applicability of domestic mountain expressways is poor,so more suitable machine learning algorithms are used for research.This paper selects the US SR-12 W as the research road section,and analyzes the factors that affect the occurrence of traffic accidents through the statistical integration of the accidents over the years,that is,traffic flow parameters,weather parameters and traffic composition.Through the selection of the detector traffic flow parameters,time period and bad weather parameters,the initial 57 characteristic variables were determined.Random Forest method is selected to rank the importance of high-dimensional feature variables.The logistic regression model was used to establish a real-time accident prediction model for mountain expressways.The input variables,optimal thresholds and optimal parameters suitable for the mountain expressway model were studied.Through collinearity diagnosis,12 characteristic variables were finally determined as input variables of the prediction model.The images of the four indicators varying with the thresholds were given,and the most The optimal threshold is 0.24,and the parameters of the model are optimized by the gradient descent method.Using Python as a means,a random forest feature variable screening model and a real-time accident risk prediction model were established.The accuracy of the prediction model was83%.On the basis of the above model,a joint AEB simulation platform of CarSim and Simulink was built,and the prediction model was compared and verified.Aiming at the adaptability of the model,the matching degree test was carried out on the domestic mountainous Zhangshi Expressway.This model also has high applicability in mountain expressways in China.