Analyzing Importance Measure methodologies for integrated Probabilistic Risk Assessment in Nuclear Power Plants

Importance Measures (IMs) are used to rank the risk contributing factors in Probabilistic Risk Assessment (PRA). In this paper, existing IM methodologies are analyzed in order to select the most suitable IM for an Integrated PRA (IPRA) of Nuclear Power Plants. In IPRA, the classical PRA of the plant is used, but specific areas of concern (e.g., fire, GSI-191, organizational factors, and seismic) are modeled in a simulation-based module (separate from PRA) and the module is then linked to the classical PRA of the plant. The IPRA, with respect to modeling techniques, bridges the classical PRA and simulation-based/dynamic PRA. This paper compares the local and Global Importance Measure (GIM) methodologies and explains the importance of GIM for IPRA. It also demonstrates the application of GIM methodologies to illustrative examples and, after comparing the results, selects the CDF-based sensitivity indicator (S_i ^(CDF)) as an appropriate moment-independent GIM for IPRA. The results demonstrate that, because of the complexity and nonlinearity of IPRA frameworks, S_i^(CDF) is the best method to accurately rank the risk contributors. S_i^(CDF) can capture three key features: (1) distribution of input parameters, (2) interactions among input parameters, and (3) distribution of the model output.