中国安全科学学报 ›› 2023, Vol. 33 ›› Issue (10): 79-85.doi: 10.16265/j.cnki.issn1003-3033.2023.10.1924

• 安全工程技术 • 上一篇    下一篇

基于无监督LDA的水电工程施工安全事故致因分析

陈述1,2(), 孙孟文2, 陈云1,2,**(), 聂本武2,3, 李智4, 刘文濯5   

  1. 1 三峡大学 水电工程施工与管理湖北省重点实验室,湖北 宜昌 443002
    2 三峡大学 水利与环境学院,湖北 宜昌 443002
    3 国家能源投资集团有限责任公司 金沙江分公司,四川 成都 610000
    4 中国长江三峡集团有限公司,湖北 武汉 430010
    5 三峡大学 经济与管理学院,湖北 宜昌 443002
  • 收稿日期:2023-04-20 修回日期:2023-07-25 出版日期:2023-11-24
  • 通讯作者:
    **陈云(1993—),男,湖北枝江人,博士,讲师,主要从事安全管理方面的研究。E-mail:
  • 作者简介:

    陈述 (1986—),男,湖北英山人,博士,教授,主要从事水电工程施工安全管理研究。E-mail:

  • 基金资助:
    国家自然科学基金资助(52079073); 国家自然科学基金资助(52209163)

Causal analysis of construction safety accidents in hydropower projects based on unsupervised LDA

CHEN Shu1,2(), SUN Mengwen2, CHEN Yun1,2,**(), NIE Benwu2,3, LI Zhi4, LIU Wenzhuo5   

  1. 1 Hubei Key Laboratory of Construction and Management in Hydropower Engineering, China Three Gorges University, Yichang Hubei 443002, China
    2 College of Hydraulic & Environmental Engineering, China Three Gorges University, Yichang Hubei 443002, China
    3 Jinshajiang Branch, China Energy Investment Corporation, Chengdu Sichuan 610000, China
    4 China Three Gorges Corporation, Wuhan Hubei 430010, China
    5 College of Economic and Management, China Three Gorges University, Yichang Hubei 443002, China
  • Received:2023-04-20 Revised:2023-07-25 Published:2023-11-24

摘要:

为实现水电工程施工安全事故报告中致因的智能挖掘,首先,利用Jieba库分词处理 1 206 条事故分析报告,提出事故分析文本词频-逆文档频率(TF-IDF)关键词处理算法,确定词频权重并构建事故文本词向量;然后,基于TF-IDF特征,训练无监督隐含狄利克雷分布(LDA)主题模型,提取事故主题及主题词;最后,对主题词进行社会网络分析,揭示事故要素间的潜在关系,智能输出水电工程施工安全事故成因。结果表明:LDA主题模型能快速挖掘出大量有效事故数据信息,并计算出安全意识、事故隐患、违章行为等5个事故主题。致因自动分析结果显示,违规违章操作、未掌握安全操作技术、材料设备问题、违反施工程序、作业环境条件不良是导致水电工程施工安全事故的最主要原因。加强施工人员的行为监管,提高事故主要致因的预防能力,有助于提升水电工程施工安全管控水平。

关键词: 水电工程, 施工安全事故, 无监督隐含狄利克雷分布(LDA)主题模型, 事故致因, 社会网络分析, 因子分析

Abstract:

To achieve the intelligent mining of causative factors in hydropower construction safety accident reports, 1 206 accident analysis reports were processed by using Jieba segmentation. Also, the TF-IDF (Term Frequency-Inverse Document Frequency) keyword processing algorithm for accident analysis text was proposed, which was used to determine word frequency weight and construct word vector of accident text. Furthermore, based on the TF-IDF features, the unsupervised LDA topic model was trained to extract accident topics and topic words. Finally, social network analysis was performed on the topic words to reveal the potential relationships among accidents elements and intelligently output the causes of hydropower engineering construction safety accidents. The results show that the LDA theme model can quickly dig out multiple accident data information effectively, and calculate five accident themes such as safety awareness, accident hazards and violation behaviors. Besides, the results of automatic cause analysis indicate that the most important causes of construction safety accidents in hydropower projects are violation of rules and regulations, failure to master safe operation techniques, material and equipment problems, violation of construction procedures and poor working environment. The behavior supervision of construction personnel should be strengthened, to improve the prevention ability of the main causes of accidents, which can effectively improve the safety control level of hydropower project construction.

Key words: ydropower engineering, construction safety accident, unsupervised latent Dirichlet allocation (LDA) topic model, accident causation, social network analysis, factor analysis