Research On Exploratory Visual Analysis Method Of Traffic Accident Cause Mode

Traffic accidents pose great threats and challenges to road traffic safety.Analyzing the cause of accidents has played a positive role in reducing the number of accidents and reducing the losses caused by accidents.Due to the inherent complexity of road traffic systems,common automated algorithms are not flexible enough.Visual analysis introduces the expertise of domain experts into the analysis process in an interactive way,and improves the quality and efficiency of analysis results by combining machine intelligence with human intelligence.Therefore,the dissertation proposes a method of combining the domain knowledge of road traffic experts and machine intelligence through visual analysis methods,and iteratively optimizing the analysis results,and use the basic traffic database of Yunnan Province from 2006 to 2018 to develop the accident cause model.The main research contents include:(1)Research on the exploratory division method of homogeneous high-dimensional data: The cause of the accident belongs to high-dimensional data.Aiming at the heterogeneity of high-dimensional data,this method proposes to first divide the data through the clustering method,and then according to the feedback information of the analysts Iteratively optimize the homogeneity data division framework of the division results,and then provide corresponding solutions to the three key issues of optimized division information in the framework,rule subset screening recommendations,and how to verify analysts’ conjectures.Finally,multi-view collaboration is used to assist analysts in freely exploring data.This method can be used not only to mine the hidden local cause patterns in accidents,but also to discover the local patterns of other highdimensional data.(2)Research on the exploratory division method of continuous attributes in numerical correlation analysis: This method converts the correlation between driver’s historical violation records and accidents into numerical correlation analysis problems,and proposes a rough division on the key issue of continuous attribute division.And carry out the association analysis,and then the analysts are free to explore the association rules and give partition suggestions,and then further partition according to the iterative partition method of the suggestion.A subregion generation algorithm based on constraint optimization is proposed to recommend meaningful candidate subregions.In response to the conflicts suggested by the analysts,a regional integration algorithm is proposed to resolve conflicts.Finally,this dissertation provides a set of visual analysis system including multi-view linkage to allow analysts to observe and analyze the rules from different angles,so as to discover more valuable information and achieve further optimization of the partition results.