Methods On Highway Accident Prone Sections Identification And Spatiotemporal Correlation Analysis

Highway traffic accidents are generally more serious than urban areas due to the fast speed of vehicles on the highway and the difficulty of manual intervention of traffic flow,etc.To a large extent,this hinders the orderly operation of highways traffic safety,and even threatens the safety of people’s lives and property.Therefore,it has always been a crucial concern in the field of traffic safety to investigate the spatio-temporal distribution of traffic accident-prone sections of highways and spatio-temporal correlation law of various factors.The identification method of traffic accident-prone sections based on accident statistics may have the modifiable areal unit problem(MAUP)in geography,the size of statistical unit directly affects the identification results and accuracy of accident-prone sections.Additionally,traffic accident is a typical spatiotemporal event,which often contains specific spatio-temporal laws.Based on these,this thesis aims at identifying the traffic accident-prone sections and exploring the accident occurrence mechanism and spatio-temporal correlation law,by mining the historical traffic accident data of highways,so as to provide auxiliary support for the active risk assessment of highway traffic accidents and intelligent control.The followings are the main content:(1)This thesis introduces the data preprocessing method of accurate analysis of traffic accident location and classification of accident point data attributes based on semantic matching,and the two parts of the accurate position analysis method,the location coordinate calculation of the accident point and the coordinate error correction are described in detail.Finally,the spatio-temporal and attribute aggregation characteristics of traffic accident data are preliminarily analyzed from two aspects of spatial-temporal distribution and accident attribute statistics.(2)A traffic accident-prone section identification method based on spatio-temporal density clustering is proposed.This method improves the traditional DBSCAN spatial clustering algorithm,and proposes the calculation method of spatio-temporal proximity distance of accident points based on angle division.Considering the influence of accident severity on traffic accidents,it divides the accident severity into different levels according to the number of deaths,the number of light injuries and the number of serious injuries,and gives different weights to different levels of accident points,so as to redefine the calculation of the minimum number of adjacent points in the clustering algorithm.According to the improved DBSCAN density clustering method,the accident-prone sections of historical accident data in Hunan Province are identified.The experimental results are analyzed from two aspects of the overall and typical roads,and the reliability of this method is verified through case study.(3)The attribute table of highway traffic accident based on multi-source data fusion is constructed.The attribute factors are selected from the space and time of the accident,accident participants,vehicles,roads and environment for discretization.The Bayesian network models of the whole test area and the typical accident prone sections identified in this thesis are constructed respectively,and the influence of different accident-causing factors on the accident severity is analyzed from the perspectives of full-attribute and multi-attribute.By comparing the research conclusions of the whole region and the typical traffic accident-prone sections in Hunan Province,there is a nonnegligible spatio-temporal heterogeneity between them.Therefore,taking appropriate measures to prevent specific roads in advance can effectively reduce the probability of serious traffic accidents.