Analysis Of Traffic Accident Injury Severity On Freeway Based On Improved Logit Model

Freeway traffic crashes is one of the major risk to personal and property safety,which needs to be solved immediately.The key to solve the problem is to study the accident injury severity and take corresponding measures to reduce accident casualties according to the regularity of injury severity reflected by the data.In this paper,based on the basic discrete choice model(Multinomial Logit model),a combination injury severity analysis prediction model combining meso and micro level analysis is established,the Latent Mixed Logit model(referred to as the "improved model")which takes heterogeneity and difference between parameter groups into account.In order to ensure that the model captures the difference between single-vehicle accidents and multi-vehicle accidents,this paper analyzes single-vehicle accidents and multi-vehicle accidents respectively.Firstly,the basic theory and principle of discrete choice model of traffic accident injury severity are compared and analyzed,and Multinomial Logit model is selected as the basic method of accident injury severity analysis.Multinomial Logit models of accident injury severity were built based on the freeway accident data of Heilongjiang province from 2008 to 2017.Secondly,in order to describe the heterogeneity of the influence of various factors on the injury severity of the accident,the Mixed Logit model was constructed by introducing random parameters on the basis of Multinomial Logit models,and the simulated maximum likelihood estimation method was used to estimate the parameters of the Mixed Logit model.In order to reflect the difference between parameter groups in Multinomial Logit models,a latent class Logit model was constructed,and the maximum likelihood estimation method was used to solve the model.Then,in order to comprehensively analyze the heterogeneity and the difference between parameter groups,the Mixed Logit model and the Latent Class Logit model were combined to construct the Latent Mixed Logit model.In order to improve the efficiency and accuracy of parameter estimation,the Markov Chain Monte Carlo complete Bayesian parameter estimation method was introduced to solve the parameters in the improved model and evaluate the improved effect of model fitting and prediction.Finally,K-fold cross-validation was used to verify the generalization ability of each model,and key factors affecting accident injury severity were analyzed according to parameter estimation results,and targeted safety improvement measures were proposed based on sensitivity analysis.The results show that the improved model has the advantages of both random parameter model and Latent Class model,which can fully describe the heterogeneity and effectively describe the difference between parameter groups.The influence factors of injury severity level of multi-vehicle accident and single-vehicle accident are obviously different,which should be considered comprehensively in the process of improving traffic safety.