Study Of Several Problems Of Human Reliability Analysis In Nuclear Power Plant

Human reliability analysis (HRA) is an essential part of probabilistic safety assessment of nuclear power plant (NPP). Good human reliability analysis leads to more accurate assessment of system safety; it can also prevent and reduce human error in large, complex human-machine system. However, the lack of understanding of human cognitive mechanism and the lack of of human error data has been plagued HRA research and application for years. But human reliability analysis has very urgent needs. Therefore, in this context, the study appears to be more open and more exploratory. We have long been engaged in human reliability research and application, including HRA in several domestic nuclear power plants. We encountered many problems in our work; we also formed some new ideas and found some new solutions in handling and solving these problems. The major research work done as follows.Firstly, we analyzed the behavior of operators in NPP systematically, introduced the Unified Modeling Language for building workflow model of the behavior of operators. Workflow model of operators can accurately reflect the property of behavior and internal constraints. Moreover, we studied simulation method of human reliability, including some key problems in human reliability simulation, such as simulation modeling, simulation algorithm design, random numbers generation, error analysis, and simulation size decision. An example of human reliability simulation is presented referring to a widely used HRA method. Simulation results show that simulation is more effective in dealing with time interface, error propagation and sensitivity analysis. Simulation makes it possible to study human performance from different angles.Secondly, we proposed a hierarchical causal model of human reliability. Bayesian network is used to build the causal link between different factors. Hierarchical structure is used in the causal model, which leads to a clear and simplified relationship between performance influence factors. Computational cost is also reduced largely. In the quantification of the model, we designed a simplified method to calculating the impact of factors; weights of factors were converted to conditional probability. We also proposed a novel variable elimination algorithm to improve reasoning performance of Bayesian network. In the novel algorithm, the added edges and the removed edges during elimination are considered; some methods of reducing graph complexity and controlling elimination cost are used to improve performance. Simulation results show that the new algorithm is effective.Thirdly, in order to solve the problem of lack of data in human reliability analysis, a human reliability database system was developed. A hierarchic architecture of human reliability data was built, which included two parts, one was human error mode classification architecture, and the other was performance shape factor classification architecture. Based on theoretical analysis, a human reliability database system was implemented; useful data was stored in this database. The extrapolation module of this system can support qualitative and quantitative analysis of human reliability. This system is useful for data collecting and data analysis of human reliability, which can provide basic support for future research.