摘要
为预防和减少建筑施工事故,从人员、管理、外在环境等角度,利用机器学习算法辨识事故发生的风险倾向。首先,针对人、管、环3因素,分别设计并检验施工人员过度自信、组织管理氛围以及施工环境危险水平测量量表;其次,根据量表得分,采用K-means算法划分出不同事故风险倾向类型,提出综合风险因子判断聚类结果;最后,运用主成分分析(PCA)、Borderline-SMOTE以及LightGBM算法逐步构建施工事故风险倾向判别模型,采用k折交叉验证法评估模型稳定性。结果表明:模型预测准确率达90.83%,能够有效辨识建筑事故的风险水平,并将事故风险分为低风险、中等风险、高风险3种倾向类型。
In order to prevent and reduce construction accidents,a machine learning algorithm was used to identify the risk propensity of accidents from the perspectives of humans,management,and external environments.Firstly,according to the three factors of humans,management,and environments,the scales for measuring the overconfidence of construction personnel,organizational management atmosphere,and construction environment risk levels were designed and tested respectively.Secondly,according to the score of the scale,the K-means algorithm was used to classify different types of accident risk propensity,and comprehensive risk factors were proposed to judge the clustering results.Finally,principal component analysis,Borderline-Smote,and LightGBM algorithms were used to establish the risk propensity discrimination model of construction accidents,and the stability of the model was evaluated by the k-fold cross-validation method.The results show that the prediction accuracy rate of the model is 90.83%.Therefore,the model can effectively identify the risk level of construction accidents,and the accident risks can be divided into three types:low-risk propensity,medium-risk propensity,and high-risk propensity.
作者
熊坚
彭一鸣
陈瑶
宋致科
蔡晶
XIONG Jian;PENG Yiming;CHEN Yao;SONG Zhike;CAI Jing(Faculty of Transportation Engineering,Kunming University of Science and Technology,Kunming Yunnan 650500,China;Yunnan Transportation Research Institute Co.,Ltd.,Kunming Yunnan 650011,China;CITIC Construction Co.,Ltd.,Beijing 100027,China)
出处
《中国安全科学学报》
CAS
CSCD
北大核心
2023年第S01期69-75,共7页
China Safety Science Journal
基金
云南省交通厅科技创新及示范项目(HZ2021X0302A)。
关键词
建筑施工事故
风险倾向
过度自信
组织管理氛围
施工环境
LightGBM算法
construction accidents
risk propensity
overconfidence
organizational management atmosphere
construction environment
LightGBM algorithm
