炼化企业安全隐患分析中LDA模型参数敏感性研究OA
Sensitivity Study on Parameters of LDA Model in Safety Hazard Analysis of Refining and Chemical Enterprises
聚焦炼化企业双重预防机制数字化系统中的非结构化隐患数据,提出基于文本挖掘与隐含狄利克雷分布(Latent Dirichlet Allocation,LDA)主题模型的隐患分析方法.以某石化公司 2022-2023 年10 555 条隐患数据为样本,通过构建专业词典库提升分词精度,运用词频-逆文本频率指数(Term Fre-quency-Inverse Document Frequency,TF-IDF)算法筛选核心关键词,并通过随机搜索优化 LDA 模型参数,最终挖掘出5 类基础隐患主题及3 类细分主题.研究表明,该方法可有效揭示高频隐患特征,为企业精准制定隐患排查策略提供技术支撑.
Focusing on the unstructured hazard data within the digital systems of the dual prevention mecha-nism in refining and chemical enterprises,this study proposes a hazard analysis method based on text mining and the Latent Dirichlet Allocation(LDA)topic model.A total of 10 555 hazard records collected from a Petro-chemical company between 2022 and 2023 were utilized as the research sample.To enhance the accuracy of word segmentation,a specialized safety lexicon was constructed.Subsequently,the Term Frequency-Inverse Document Frequency(TF-IDF)algorithm was applied to extract core keywords,and a random search method was employed to optimize the parameters of the LDA model.Consequently,five fundamental hazard topics and three detailed sub-topics were successfully extracted.The results indicated that this method can effectively re-veal the characteristics of high-frequency hazards.Therefore,these findings provide robust technical support for enterprises to precisely formulate targeted hazard inspection strategies.
厉建祥;徐杨;尹志攀;李千登
中石化安全工程研究院有限公司,山东 青岛 266104||中石化管理体系认证(青岛)有限公司,山东 青岛 266104中石化安全工程研究院有限公司,山东 青岛 266104||中石化管理体系认证(青岛)有限公司,山东 青岛 266104中国石油化工股份有限公司金陵分公司,江苏 南京 210033中石化安全工程研究院有限公司,山东 青岛 266104||中石化管理体系认证(青岛)有限公司,山东 青岛 266104
资源环境
炼化企业双重预防机制文本挖掘LDA主题模型隐患分析参数优化
refining and chemical enterprisesdual prevention mechanismtext miningLatent Dirichlet Allocation(LDA)topic modelhidden hazard analysisparameter optimization
《安全、健康和环境》 2026 (6)
82-87,6
中国石化化工事业部项目(CHG26007),流程驱动的炼化基层岗位HSE职责落实与评价技术研究.
评论