【作者】 蒋英杰；

【导师】 谢红卫；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2012， 博士

【摘要】 人为差错严重影响人机系统的可靠性和安全性，已经成为导致事故的主要原因之一，研究人为差错发生的特点及规律，并在此基础上设计人为差错的管理措施具有十分重要的意义。论文以工程实践对人因可靠性分析方法的现实需求为牵引，以推动人因可靠性分析研究的发展为导向，在人因可靠性分析的理论和方法方面开展了一系列深入的研究工作。具体的研究思路为：在已有成果的基础上，系统性地研究了人的认知行为描述、情景环境的表征、人为差错的辨识和人为差错的概率量化等4个人因可靠性分析的核心内容，提出了一系列具有规范性或创新性的方法，并通过工程应用验证了方法的工程适用性和有效性。论文的主要研究内容和成果如下：（1）人的认知行为描述通过分析比较现有的认知模型，选择SRK模型来描述人的认知行为过程，并以此作为论文后续研究的心理学基础。为了提高SRK模型的理论合理性，更新了SRK模型的解释。考虑到行为模式决定因素的状态可能存在不确定性，分别使用模糊逻辑和概率的方法，构建了静态和动态两种情景环境中行为模式的确定方法，使得行为模式的确定方法更具适用性。（2）情景环境的表征以人机交互过程为范本，分析得到了包含6个方面、3个层次、38个具体元素的行为形成因子分类，将情景环境的内涵描述得更加全面、细致和规范。根据SRK模型中3种行为模式的特点，对行为形成因子作了再分类，进一步得到了与3种行为模式对应的主要和次要行为形成因子。借鉴和适度规范了行为形成因子评分的专家判断法，有助于提高行为形成因子评分结果的有效性。通过分析人为差错数据，提出了基于关联规则挖掘的行为形成因子权重确定方法，提高了行为形成因子权重确定方法的有效性和工程适用性。最后，给出了综合评价情景环境的基本步骤，对行为形成因子的选择、评分和权重确定中所需要注意的问题，都进行了充分的讨论。（3）人为差错的辨识首先，改进和完善了外部人为差错的辨识方法。通过分析和归纳，得到了外部人为差错辨识方法的基本框架，以“执行差错基本分类”为分类框架，提出了层次性、完备性和细致性兼顾的外部人为差错分类方法，通过在分类的基础上设计“问题”，构建了基于“问题”引导的外部人为差错辨识方式。然后，创新性地着重研究了内部人为差错的辨识方法。分析构建了内部人为差错辨识方法的基本框架，分析了认知模式差错、认知功能差错和认知行为差错3者之间的关系，提出了基于SRK模型的内部人为差错分类方法，通过在分类的基础上设计“问题”，构建了基于“问题”引导的内部人为差错辨识方式，弥补了现有方法的不足。另外，通过分析行为形成因子与认知行为差错模式之间的关联关系，提出了查找认知行为差错模式成因的辅助方法。最后，将外部人为差错的辨识和内部人为差错的辨识有机结合，给出了人因可靠性工程中全面开展人为差错辨识工作的流程图。（4）人为差错的概率量化探索性地建立了人为差错概率量化的基本框架。分析了情景环境、行为模式和人为差错概率3者之间的映射关系，分别设计了静态和动态两种情景环境中人为差错概率的预测方法，提高了人为差错概率预测方法的工程适用性。分析了人为差错数据的来源，分别设计了基于Bayes基本法和基于Bayes信息融合的人为差错概率的修正方法，用于提高人为差错概率量化结果的有效性。（5）工程应用以研究成果为基础，形成了人因可靠性分析的工程方案，详细分析了汽车驾驶过程中驾驶员的认知行为描述、驾驶过程情景环境的表征、驾驶员人为差错的辨识以及驾驶员人为差错的概率量化等4个人因可靠性分析过程的主要环节。分析结果表明，所提出的方法弥补了现有方法存在的一些缺陷和不足，具有较好的工程适应性和有效性，能够为人因可靠性工程的有效开展提供帮助和指导。更多还原

【Abstract】 Human errors have more negative effect on the reliability and safety ofhuman-machine systems than machine faults nowadays, and have been recognized asthe main cause of incidents or accidents. Therefore, it is very important to researchhuman errors in order to control them in an effective way. Driven by the requirementsfrom engineering and in order to promote the development of human reliability analysis,this dissertation researches four key contents of human reliability analysis i.e. thehuman cognitive behavior description, the scenario representation, the human erroridentification and the human error probability quantification, and therefore proposes aseries of techniques for standardization or innovation. An engineering case study isprovided, which illustrates the applicability and validity of the proposed techniques. Theoutline of research works and achievements are as follows.(1) Human cognitive behavior descriptionThe available cognitive models are collected and compared. By comparison, SRKmodel is selected as the psychological base of human cognitive behavior description inthis dissertation. The explanations of SRK (Skill, Rule and Knowledge) model areupdated to improve its theoretical rationality. Considering that there are uncertaintiesamong the states of the behavior mode determinative factors, two techniques, based onfuzzy logic system and probability respectively, are proposed to determine behaviormode in the static and the dynamic scenario.(2) Scenario representationThe dissertation chooses the human-machine interaction process as the typicalparadigm. Performance shaping factors, taken as the representatives of scenario, arecategorized into six aspects, three levels and thirty-eight factors, which can describe thescenario in a more complete, delicate and normalized way. According to the propertiesof the three behavior modes in SRK model, performance shaping factors are categorizedfurther, and the dominant and subordinate ones corresponding to each behavior modeare discriminated. The technique to rate performance shaping factors by expertjudgement is employed and normalized to some extent, so that the validity of the ratingresults can be improved. By analyzing human error data, the technique based onassociation rule mining is proposed to weigh performance shaping factors, which canimprove the validity and applicability of the current techniques. Based on the proposedtechniques, the basic scheme to evaluate the whole scenario is given, and the problemsthat need be noticed when implementing the scheme are discussed in detail.(3) Human error identificationFirst, the external human error identification is improved. By analyzing andsummarizing, the basic framework for the external human error identification is provided. Taking “the basic taxonomy of execution errors” as the taxonomic framework,the taxonomy of external human errors is provided. Base on the taxonomy, a series of“questions” are designed and the “question-answer” identification manner is proposedaccordingly. Second, the internal human error identification is researched innovativelyas the main topic. The basic framework for the internal human error identification isconstructed. The correlations of the errors of cognitive modes, cognitive functions andcognitive behaviors are analyzed, and the taxonomy of internal human errors based onSRK model is proposed. With the provision of the taxonomy, a series of “questions” aredesigned and the “question-answer” identification manner is constructed accordingly.Besides, the correlation between performance shaping factors and human cognitivebehavior error modes are analyzed, and the technique to facilitate the causes-findingprocess of human cognitive behavior error modes is proposed. At last, the external andthe internal human error identification techniques are combined, and therefore theflowchart for the full-scope application of human error identification is provided.(4) Human error probability quantificationThe basic framework for the human error probability quantification is provided. Asto the human error probability prediction, the correlations among scenario, behaviormode and human error probability are analyzed and constructed, and the quantificationtechniques that can be used in the static and the dynamic scenario respectively areprovided, which can improve the applicability of the current techniques. As to thehuman error probability amendment, the multiple sources of human error data areanalyzed, and two techniques based on Bayesian basic method and Bayesianinformation fusion method are proposed. They can improve the validity of the results ofthe human error probability quantification.(5) Engineering applicationBased on the techniques proposed in the dissertation, an engineering solution forhuman reliability analysis is formed, and it is applied in and demonstrated by the casestudy of human errors analysis on car drivers. The cognitive behavior description,scenario representation, human error identification and human error quantification ofcar drivers are analyzed in detail. The results show that the proposed techniques canovercome some shortcomings of the existed techniques and can provide betterapplicability and validity. It is a useful solution for the human reliability engineering.更多还原