期刊文献+

语义识别驱动的化工泄漏事故事前预防研究

Research on the prevention of chemical leakage accidents driven by semantic recognition
下载PDF
导出
摘要 化工泄漏事故报告蕴含事故信息量大,但利用度低,仅依赖传统的事故分析理论和方法对事故后果进行分析统计难以实现事前预防、控制损失最小化的目的,因此,构建了语义识别驱动的化工泄漏事故事前预防研究框架,基于潜在狄利克雷分配(Latent Dirichlet Allocation,LDA)主题模型提取化工泄漏事故致因主题及关键词,利用关键词共现网络分析进行致因中心性和关联度分析,使用因子分析进行致因影响因子的计算,实现了对化工泄漏事故报告潜在信息的挖掘和有效分析。结果表明:通过LDA模型可以计算得到化工泄漏事故致因主题,得出安全意识缺失、物料逸出、设备故障等5个聚类;基于改进点互信息(Pointwise Mutual Information,PMI)的关键词共现网络可以得到事故的关键致因、环节、场所和事故类型,其中最重要且关联度较高的致因是人员操作不当和现场管理不力;最后,通过因子分析得到影响后果最严重的致因是危险作业环境,其次是违规操作或操作不当。提出的研究框架在更深入挖掘利用海量事故致因信息的同时,减少了事故致因评价指标的主观性,为结构复杂、非单一标准的事故报告文本信息提取提供了新的思路,同时将语义识别拓展到化工泄漏事故预防领域,有助于化工泄漏事故的风险识别、预测与防控。 Chemical leakage incident reports contain a wealth of information;however,their utilization remains low.Solely relying on traditional accident analysis theories and methods for post-incident consequence analysis and statistics is insufficient for achieving proactive prevention and minimizing losses.This study establishes a semantic recognition-driven research framework for the proactive prevention of chemical leakage incidents.The Latent Dirichlet Allocation(LDA)topic model is employed to extract causative themes and keywords.The results demonstrate that LDA modeling effectively identifies the causative themes associated with chemical leakage incidents from the reports.Subsequently,keyword co-occurrence network analysis is used to examine the centrality and associative degree of the causative factors.Factor analysis is performed to calculate the impact of these elements,enabling the extraction and effective analysis of potential information from chemical leakage incident reports.This analysis categorizes the causative factors into five clusters:lack of safety awareness,material escape,equipment failure,and others.Using the improved Pointwise Mutual Information(PMI)-based keyword co-occurrence network,critical causative factors,stages,locations,and types of accidents are identified.The analysis reveals that the most significant and strongly associated causative agents are improper personnel operations and inadequate on-site management.Factor analysis further reveals that the most significant contributory cause is hazardous working environments,followed by violations and improper operations.The proposed research framework substantially enhances the extraction and utilization of extensive causative information from accident reports,thereby reducing the subjectivity inherent in evaluations of accident causation.By offering new methods for extracting information from complex,non-standardized accident report texts,this study expands the application of semantic recognition in the field of chemical leakage accident prevention.This advancement is essential for enhancing the identification,prediction,and management of risks associated with chemical leakage incidents.In conclusion,the integration of LDA,keyword co-occurrence network analysis,and factor analysis within this framework provides a systematic,data-driven approach to the proactive prevention and comprehensive analysis of chemical leakage incidents.This approach demonstrates significant practical applicability and holds great potential for enhancing safety management practices.
作者 刘勤明 董宏霖 孔得朝 LIU Qinming;DONG Honglin;KONG Dechao(Business School,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Intelligent Emergency Management,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处 《安全与环境学报》 CAS CSCD 北大核心 2024年第12期4734-4742,共9页 Journal of Safety and Environment
基金 国家重点研发计划青年科学家项目(2021YFF0900400) 上海市2021度“科技创新行动计划”宝山转型发展科技专项项目(21SQBS01404)。
关键词 安全社会工程 化工事故 文本挖掘 语义识别 词频逆文档频率算法 潜在狄利克雷分配主题模型 safety social engineering chemical accident text mining semantic recognition term frequency-inverse document frequency algorithm latent dirichlet allocation topic model
  • 引文网络
  • 相关文献

参考文献9

二级参考文献66

共引文献240

;
使用帮助 返回顶部