摘要
地铁施工事故具有易发性且事故类型复杂多样,针对现有地铁施工事故分析方法多依赖于专家主观经验构建,且仅有较少方法对事故报告信息加以利用的问题,提出一种基于NBN-EM的地铁施工事故致因分析模型。首先,以搜集的2010—2021年间的223起事故报告为数据来源,采用统计学方法提取及筛选风险因素,进一步归纳建立事故致因分析的指标体系;其次,采用改进的朴素贝叶斯网络构建风险因素指标与事故类型关系的图形结构,同时分层随机抽样80%的数据为训练样本,借助EM算法和Netica软件进行数据学习,确定各节点的先验概率和条件概率参数;最后,通过贝叶斯网络推理和敏感性分析得到不同类型事故的关键致因排序,不同视角下的情景分析对风险因素组合作用下的事故发生概率和风险源识别进行了确定。研究结果表明:施工工法、施工方案内容、安全隐患排查分别为三个维度上造成事故发生的最重要因素,不同类型事故的关键风险因素具有差异性,应区别预控,模型测试验证方法的有效性,平均正确率为84.55%。
Metro construction accidents are prone to occur and the types of accidents are complex.Aiming at the problem that the existing metro construction accident analysis methods mostly rely on expert subjective experience,and only a few methods make use of accident report information,this paper proposes an analysis model for the causes of metro construction accidents based on NBN-EM.First of all,223 accident reports collected from 2010 to 2021 are used as data sources,and statistical methods are used to extract and screen risk factors and further generalize them to establish an indicator system for accident causation analysis.Secondly,an improved Naive Bayesian network is used to construct the graphical structure of the relationship between risk factor indicators and accident types,while a stratified random sampling of 80%of the data is used as the training sample,and data learning is carried out with the help of EM algorithm and Netica software to determine the prior probability and conditional probability parameters of each node.Finally,Bayesian network inference and sensitivity analysis are employed to rank the key causal factors of different types of accidents,and scenario analysis from different perspectives determines the accident probability and risk source identification under the combination of risk factors.The research results show that the construction method,construction plan content and safety hazard investigation are the most important factors that cause accidents in the three dimensions respectively.The key risk factors for different types of accidents are differentiated and should be pre-controlled differently.The model testing verifies the validity of the method with an average correct rate of 84.55%.
作者
申建红
刘树鹏
SHEN Jianhong;LIU Shupeng(College of Management Engineering,Qingdao University of Technology,Qingdao 266520,China;Research Institute of Construction Credit and Risk Management,Qingdao University of Technology,Qingdao 266520,China)
出处
《铁道标准设计》
北大核心
2024年第6期171-179,共9页
Railway Standard Design
基金
国家自然科学基金项目(71471094)
国家住房和城乡建设部科学技术计划项目(2022-R-048)。
