A Critical Review and Benchmark Study of Dependency Modeling for Seismic Probabilistic Risk Assessment in the Nuclear Power Industry

The risk significance of seismic dependencies has received growing attention from the nuclear power industry, particularly following the Fukushima disaster in 2011. As a result, many efforts have been devoted to dependency modeling for seismic probabilistic risk assessment (PRA) of nuclear power plants. However, there is still no consensus on a unified method for seismic dependency modeling. This paper aims to review and benchmark the state-of-the-art methods for seismic dependency modeling. This is accomplished by investigating an asymmetric system consisting of three safety -related components. Notably, we illustrate the application of four leading dependency modeling methods: full Monte Carlo simulation (MCS), Bayesian network, ReedMcCann, and COREX methods. The paper compares and contrasts these methods' performance from three perspectives: the precision of estimates within analytical lower and upper bounds; the proper differentiation between independent and dependent scenarios; and the degree of conservatism in dealing with parallel systems. The results show that the Reed -McCann method has issues in quantifying the effect of seismic dependencies and provides considerable conservatism; the Bayesian network -based method performs poorly and underestimates risk; the COREX method performs satisfactorily to quantify seismic dependencies with acceptable conservatism and potentially offers a balance between accuracy and scalability in large-scale seismic PRA.