Real-time Highway Accident Based On PSO-BP Predictive Model

The issue of highway traffic safety is still a concern.Establishing a real-time prediction model for highway accidents based on existing accident data is of great significance in reducing casualties caused by accidents and vehicle delays after accidents.The existing research results mainly focus on the real-time accident prediction and the relationship between the influencing factors and the severity of the accident.Research on the real-time prediction of the severity of highway traffic accidents still needs to be studied.This paper selects the American highway I405 N as the research object,after sorting out a total of 95 characteristic variables such as traffic flow data,climate data and other influencing factors,a total of 12 characteristic variables are selected as input variables based on the characteristic variable screening method of random forest,including accidental sections The downstream traffic volume change degree,vehicle type,time period,weather and other influencing factors,using particle swarm optimization BP network to establish a real-time prediction model of the severity of highway traffic accidents.The average accuracy rate of the test group reached 84.29%,the accuracy rate of accident-free traffic conditions reached 85%,the accuracy rate of property damage accidents was 81%,the accuracy rate of personal injury accidents was 72%,and the accuracy rate of death accidents was 99%.In the same situation,it is necessary to introduce more relevant types of data to modify the prediction model.The results show the feasibility of applying particle swarm optimization BP network to the highway real-time accident severity prediction model,and the prediction accuracy and fit are significantly higher than the traditional BP model.