人因可靠性分析中的概率因果模型

Probabilistic causal model of human reliability analysis

根据人因失误的机理和特点提出了一个分层的人因可靠性概率因果模型,采用贝叶斯网建立了人因可靠性影响因素之间的因果关系。采用分层方法进行建模,充分利用了条件独立性以降低模型的复杂度,同时分层机制也符合人因可靠性的内在要求。该模型既能满足回溯型分析需求又能满足预测型分析需求。在模型的定量化方面,设计了一个简化可行的模型参数计算方法。该模型可以用于工程化的人因可靠性分析,也可扩展为人因行为理论研究模型。

A proper causal model of human reliability analysis helps to account for human response behavior and lead to a more effective quantitative analysis. In this paper, we have proposed a hierarchical causal model of human reliability in accordance with the underlying reasons of human errors, with Bayesian network being used to build the causal link of all the different factors. And, then, with the hier- archical structure adopted in the model, the factors in each level would be only affected by the factors belonging to the upper levels. In addition, since the hierarchical structure may contribute more to a clear and simplified relationship between the behavior influence fac- tors. Specifically speaking, human reliability is supposed to be affect- ed by some inherent factors, though such inherent factors may also be affected by some external factors. However, the external factors of different groups may prove to he independent of each other. There- fore, it is possible to reduce the computational load greatly. In the above model we have initiated, human inherent factors may include confidence and responsibility, knowledge and experience, psycholog- ical stress and working load, fatigue and so on. While confidence and responsibility are usually influenced by such external factors as safety culture, organizational management, and team collaboration, knowl- edge and experience may be influenced by each one＇ s education background, training level and employment history in similar posi- tions. Whereas psychological stress is impacted by help information, human-machine interface, and the availability of operational proce- dures, the working load may be affected by the complexity of the job assigned and the time urgency. As to fatigue, it may be caused either by physical environment or by some special physical barriers. To quantify the likely causes and effects used in our model, we have worked out a simplified method for calculating the impact of each fac- tor and transformed weights of the above factors into conditional prob- abilities. Thus, it can be seen that the causal model can be applied both for retrospective analysis and predictive analysis. Besides, a computeraided tool has also been initiated based on this model so as to facilitate the application of human reliability analysis.