Study On Key Technologies Of Traffic Safety Decision Support Based On Mass Data Of Traffic Management In Public Security Field

Security and smooth traffic is the core issue of road traffic management,and the decision support model for road traffic safety is an important means of solving this problem.The purpose is to enhance the decision-making ability of management and reduce road traffic accidents,by improving the model of accident hotspot recognition,key elements recognition,accident frequency prediction and evaluation of road traffic safety.This will provide the department of industry management with the better road traffic safety hotspots management and their distribution and evolvement rule.The study of model and method for decision of road traffic safety is of high complexity,because the mechanism of road safety incident has the outstanding characteristics such as dynamic,random and uncertainty,and the basic data in the accuracy,consistency,integrity is also different.With the help of massive business database of public security traffic management,this paper used self-organizing neural network(SOM),rough set theory(RS),and linear regression data mining theory and methods.It regards improving the decision support ability of road traffic safety as the goal.The model for identification,prediction and evaluation and the key elements of road traffic safety are studied deeply.We mainly carried out four aspects of work:1.Aiming at the demand of discretization for the algorithm of rough set and decision-making tree,this paper proposes an improved SOM clustering discrete algorithm.In the algorithm,the initial clustering is implemented by using SOM to determine the upper limit of clustering.After taking the initial cluster centers as samples,the secondary clustering is carried on by using BIRCH hierarchical clustering algorithm to solve the high number of clustering and identify the set of discrete breakpoints.At last,the nearest neighbor to the clustering center is found for each sample in the set of discrete breakpoints,which can be used as the basis of discrete tuning.This algorithm is superior to the traditional SOM clustering discrete algorithm in the number of breakpoints(silhouette coefficient increasing by 75%)and the discrete accuracy(incompatible degree more close to 0).2.In order to avoid the inherent defects that the traditional accident analysis method separates the inner relationship between accident attributes,this paper uses Z-Score method to unify target data dimension and range,and combines the cumulative variance contribution rate with principal component factor loading matrix to construct accident data principal component score set.Based on this,the average score of cluster can be obtained through clustering analysis of the comprehensive score of principal component factor by using Canopy-Means.We establish a comprehensive evaluation of principal component and link attributes combination model,and realize the analysis of target security quantitative section.3.In order to solve the problem of correlation analysis of traffic flow,road traffic accident and the linear regularity of construction of multi-dimension,this paper constructs the relation model of accident frequency and the annual average daily traffic(AADT)based on the polynomial regression technique.Then the problem of data discrete is eliminated based on the road characteristics,traffic flow data and accident frequency,combining with empirical Bayesian method.We design correction coefficient based on the observed value and the forecast value contrast to form the Negative binomial regression prediction model of traffic accident frequency based on CMF(Crash Modification Factor),which realizes the good prediction and screening of high-grade accident.4.Aiming at the current demand of network-based management for road traffic safety,the accident hazard index is taken as quantitative criteria.And the four assessment indicators for traffic safety risk are proposed,that is,accident rate CRs of millions of vehicles,safety risk index on a section highway based on CCR,accident rate SRs based on accidental severity of millions of vehicles,and safety risk index CSR based on combined influence.Then the model of decision-making tree is developed for judging the latent form of accidents and predicting the possible growing trend of the potential traffic accident in the object region.5.The paper develops the core algorithm of road traffic hotspots,key elements recognition,accident frequency prediction and evaluation of road traffic safety,and it is on the basis of overall design at the completion of the main functions of the road traffic safety decision support system.It achieves the deep integration of analysis management module,dimensional analysis module,decision-making and evaluation module functions for road traffic safety.Through the application and measurement of the provincial police traffic safety data,the effectiveness of the research results is verified,which lays the foundation for the integrated application of key technologies of road traffic safety decision-making.