Study On Prediction Method Of Freeway Accidents Based On Time Correlation

With the continuous increase of highway mileage in China,the problem of highway traffic safety has become increasingly serious.How to reduce the number of highway traffic accidents has become one of the issues of common concern.In order to analyzing the traffic safety problems of expressway,this paper analyzes the factors which affecting traffic accidents on expressways,and builds the prediction model based on time correlation.So that to provide a theoretical basis for the safety management work of highway management department.In this paper,using the measured data of the Guangzhou-Zhuhai Expressway,a Negative Binomial(NB)model and a Generalized Estimation Equation(GEE)model are constructed to predict the number of accidents,then comparing and analyzing the prediction results.First,based on the measured data from four aspects: people,vehicles,roads,and environment,the impact of the accident on highway accidents was analyzed.Based on the previous theoretical basis,this paper using the Pearson correlation coefficient analysis method in SPSS proposed 13 independent variables.By analyzing the correlation of these variables,11 independent variables with strong correlation were obtained.Secondly,four methods of road segmentation were proposed: static road segmentation method,slope consistency method,curve radius consistency method and orderly cluster segmentation method.Based on the previous research results about the factors which affect the highway accidents,a reasonable statistical analysis of traffic accident data with 11 independent variables according to four road segmentation methods was conducted.Then,according to the inherent characteristics of the data of the Guangzhou-Zhuhai Expressway,the model for predicting the total number of accidents based on NB model and GEE model were constructed,then comparing and analyzing the model fitting effect to summarize the features of each model.Based on the results of the elastic analysis,the significant factors which affecting the total number of accidents on the highway are determined.Finally,three predictive accuracy test indicators were selected to comparing the prediction accuracy of the NB model and the GEE model which based on four correlation matrixes.The result shows that the accident prediction based on the GEE model is more accurate than the prediction based on the NB model.The accuracy of the GEE model of the 1-AR autoregressive working correlation matrix is more accurate than that of the GEE model based on independent correlation matrix,exchangeable correlation matrix or unstructured correlation matrix.