谱聚类和Apriori算法在建筑坍塌事故致因组合分析中的应用  

Application of spectral clustering and Apriori algorithm in combination analysis of construction collapse accident causes

作　　者：李珏[1,2] 蒋敏 LI Jue;JIANG Min(School of Traffic&Transportation Engineering,Changsha University of Science&Technology,Changsha 410114,China;Intelligent Construction Operation&Maintenance Management of Transportation Infrastructure Key College Laboratory of Hunan Province,Changsha University of Science&Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学交通运输工程学院,长沙410114 [2]长沙理工大学交通基础设施智慧建造与运维管理湖南省高等学校重点实验室,长沙410114

出　　处：《安全与环境学报》2024年第2期617-625,共9页Journal of Safety and Environment

基　　金：湖南省自然科学基金项目(2021JJ30744);湖南省教育厅项目(20K011)。

摘　　要：建筑坍塌事故是人员伤亡和经济损失较大的事故类型之一。为探究建筑坍塌事故不同致因之间的关联和相互依存关系,首先,选取国内2015—2020年231份建筑坍塌事故报告作为研究对象,借助R语言平台进行文本挖掘,得到43个致因。其次,运用Python进行谱聚类,根据致因之间的关联强度对其进行聚类。最后,利用关联规则挖掘Apriori算法确定建筑坍塌事故致因之间的关键关联组合。结果表明,43个事故致因可分为5类,在每一个簇类中确定了最关键的致因组合,并提出了针对性的预防措施,为坍塌事故的预防和控制提供一种新的思路。A construction collapse accident is one type of accident with high casualties and economic losses.Previous studies used different methods to investigate how causes affect the occurrence of collapse accidents.To explore the correlation and interdependence between causes of construction collapse accidents,231 construction collapse accident reports from 2015 to 2020 were collected.The R language platform was utilized to analyze the original accident reports for sampling field,segmenting words,deprecating words,and merging word groups.Then 43 collapse accident causative items were extracted from the original characteristic items,covering all direct and indirect causes in the construction collapse accident reports.A O-1 reference matrix of 43×231 dimensions was obtained according to the text mining results.Secondly,Python was employed to perform spectral clustering analysis on 43 causative items.The optimal number of clusters k=5 was determined by comparing the performance of the contour coefficient method and the elbow method on the data set.The Principal Component Analysis(PCA)was used to downscale the high-dimensional original reference matrix and performed spectral clustering based on the strength of association between causes.Finally,the association rule mining Apriori algorithm was adopted to determine the key causal combinations in the clusters.The association rules were measured by three metrics:support,confidence,and lift.The combinations with a strong correlation in the cluster were filtered out by setting a minimum confidence threshold and a minimum lift threshold of 1.0.The results indicate that 43 accident causative items are classified into 5 categories by spectral clustering,which is different from the traditional classification of accident causes based on human,machine,environment,and management.The association rules mining algorithm is used to get the key association combinations in each cluster,which reflect the correlation among the causes.Besides,the corresponding preventive measures are put forwa

关 键 词：安全社会工程 建筑施工 坍塌事故 文本挖掘 谱聚类 APRIORI算法 

分 类 号：X947[环境科学与工程—安全科学]