A Big Data-Theoretic Approach to Quantify Organizational Failure Mechanisms in Probabilistic Risk Assessment

非技术描述福岛和卡特里娜等灾难性事件清楚地表明，为了防止复杂的技术事故和维护公共安全和健康，将故障的物理和社会原因整合到一个有凝聚力的建模框架中至关重要。在本研究中，概率风险评估（PRA）、组织行为学和信息科学与数据分析学科的专家将合作提供以下关键问题的答案：(a)什么社会和组织因素影响技术系统风险？(b)这些因素如何以及为什么影响风险？(c)它们对风险的贡献有多大？现有的PRA模型不包括组织因素的完整范围。本研究探讨了组织失效的根本原因，并对其影响技术系统绩效的路径进行建模，从而更全面地将组织失效的潜在机制纳入PRA。PRA领域在设备故障和人为错误的量化方面取得了进展，用于复杂系统的风险建模；然而，目前的组织风险贡献者缺乏可靠的数据分析，超出了安全气候和安全文化调查。本研究通过开发PRA的预测因果建模和大数据理论技术填补了这一空白，通过利用文本挖掘、数据挖掘和数据分析等技术，扩展了用于风险分析的经典数据管理方法。除了对组织科学、PRA和数据分析的科学贡献外，本研究还为监管和行业决策者提供了导致风险的重要组织因素，并导致优化决策。其他应用包括组织安全指标的实时监控、高效的安全审计、深入的根本原因分析，以及风险知情的应急准备、规划和响应。该项目的多学科方法可以作为一种教育模式，使学生能够跨越学科界限进行研究。最后，本研究是产学合作的成功范例。一家核电站已承诺参与该项目，并提供了完成研究所需的数据和信息的独特途径。拟议的方法是通用的，适用于任何高风险行业（例如航空、医疗保健、石油和天然气），并将用于改进组织的安全绩效，以保护工人、公众和环境。技术描述组织产生、处理和存储大量广泛的、非结构化的数据，作为业务活动和合规要求的结果（即，纠正措施计划、根本原因分析报告、监督和检查数据等）。本研究利用这些数据资源对组织失效机制进行量化，并将其与PRA产生的技术系统风险情景进行整合。这项研究是基于社会技术风险理论，以防止仅仅从数据知情的方法产生误导性的结果。将社会技术风险理论、系统建模和语义数据分析策略相结合，将大大提高复杂系统的风险分析能力。我们将根据以下步骤开展研究：(1)扩展社会技术风险分析（SoTeRiA）框架中的因素、子因素和因果关系；(2)开发SoTeRiA中因素、子因素及其因果关系的测量技术(例如，将文本挖掘与贝叶斯信念网络相结合；进行科学约简以识别重要因素；(3)建立动态、预测的社会技术因果建模技术，(4)进行不确定性分析，(5)进行验证和验证，(6)将定量社会技术因果模型与PRA相结合，(7)进行敏感性和重要性测度分析。作为将大数据与PRA整合的先驱研究，本研究解决并量化了社会和技术系统接口产生的风险。

Nontechnical DescriptionCatastrophic events such as Fukushima and Katrina have made it clear that integrating physical and social causes of failure into a cohesive modeling framework is critical in order to prevent complex technological accidents and to maintain public safety and health. In this research, experts in Probabilistic Risk Assessment (PRA), Organizational Behavior and Information Science and Data Analytics disciplines will collaborate to provide answers to the following key questions: (a) what social and organizational factors affect technical system risk? (b) how and why do these factors influence risk? and (c) how much do they contribute to risk? Existing PRA models do not include a complete range of organizational factors. This research investigates organizational root causes of failure and models their paths of influence on technical system performance, resulting in more comprehensive incorporation of underlying organizational failure mechanisms into PRA. The field of PRA has progressed the quantification of equipment failure and human error for modeling risk of complex systems; however, the current organizational risk contributors lack reliable data analytics that go beyond safety climate and safety culture surveys. This research fills that gap by developing predictive causal modeling and big-data theoretic technologies for PRA, expanding the classic approach of data management for risk analysis by utilizing techniques such as text mining, data mining and data analytics. In addition to scientific contributions to organizational science, PRA, and data analytics, this research provides regulatory and industry decision-makers with important organizational factors that contribute to risk and leads to optimized decision making. Other applications include real-time monitoring of organizational safety indicators, efficient safety auditing, in-depth root cause analysis, and risk-informed emergency preparedness, planning and response. The multidisciplinary approach of this project can serve as an educational model, empowering students to pursue research across disciplinary boundaries. Finally, the proposed research represents a successful model of industry-academia collaboration. A nuclear power plant has committed to this project and provides unique access to data and information necessary to complete the research. The proposed methodology is generic and applicable for any high-risk industry (e.g., aviation, healthcare, oil and gas), and will be used for the improvement of organizational safety performance in order to protect workers, the public and the environment. Technical DescriptionOrganizations produce, process and store a large volume of wide-ranging, unstructured data as a result of business activities and compliance requirements (i.e., corrective action programs, root cause analysis reports, oversight and inspection data, etc.). This research leverages those data resources for the quantification of organizational failure mechanisms and their integration with the technical system risk scenarios generated by PRA. The research is based on a socio-technical risk theory to prevent misleading results from solely data-informed approaches. Combining socio-technical risk theory, systematic modeling and semantic data analytics strategies will greatly enhance risk analysis of complex systems. We will conduct our research based on following steps: (1) Expand factors, sub-factors, and causal relationships in the Socio-Technical Risk Analysis (SoTeRiA) framework, (2) Develop measurement techniques for factors, sub-factors and their causal relationships in SoTeRiA (e.g., integrating text mining with the Bayesian Belief Network; conducting scientific reduction to identify important factors; measuring of important factors), (3) Establish a dynamic, predictive socio-technical causal modeling technique, (4) Perform uncertainty analysis, (5) Conduct verification and validation, (6) Integrate the quantitative socio-technical causal model with PRA, and (7) Conduct sensitivity and importance measure analyses. As the pioneer study on the integration of big data with PRA, this research addresses and quantifies risk emerging from the interface of social and technical systems.