认知控制模式下的CREAM方法概率量化

Quantification of human error probability of CREAM in cognitive control mode

研究了人因可靠性分析(Human Reliability Analysis,HRA)中人为差错概率的量化。首先,介绍了认知可靠性与差错分析方法(Cognitive Reliability and Error Analysis Method,CREAM)中基本法的基本理论,讨论了两种概率化认知控制模式的确定方法——贝叶斯网络法和模糊逻辑法,强调了概率化认知控制模式下量化人为差错概率的必要性。然后,在分析认知模式下人为差错概率分布的基础上,通过理论推导,构建了概率化认知控制模式下人为差错概率的量化方法。为了提高计算效率,提供了人为差错概率的蒙特卡洛仿真算法。最后,通过实例,演示了方法的使用过程,并证明了方法的有效性。

Human errors have nowadays turned to be tile main factor that may reduce the reliability and safety of human-machine system, and therefore necessary to be attached special attention to. It is for this reason that the quantification of human error probability has be- come the research topic of this paper known as a key ingredient of hu- man reliability analysis （HRA）. However, the first step for us to do here is to introduce the basic method of cognitive reliability and error analysis method （CREAM） as a kind of widely accepted HRA method as well as the basic theory it involves. And, then, we would like to introduce the steps for quantifying human error probability in details. Considering that cognitive behavior mode provided by CREAM should be continuous, we have put forward two methods for defining the probabilistic control modes by HRA practitioners, which are based on Bayesian nets and the fuzzy logic, respectively. The reason for so do- ing is that if the human error probability were not to be quantified, it would be necessary to construct a method to deal with the human error probability in probabilistic mode, which makes it necessary to apply a method for quantifying the human error probability in probabilistic control modes. In preparing for such a method, we should take the lognormal function as the probabilistic density function of human error probability in the mode and the probabilistic density function of hu- man error probability in probabilistic cognitive behavior mode as the linear combination of the functions in each cognitive behavior mode. However, the human error probability in prebabilistic mode is quanti- fied through theoretical inference. In order to heighten the efficiency of calculation, we have also applied the Monte Carlo algorithm to our work. And, last of all, the validity of the method has been demon- strated by means of a sample study to show the process of the method.