Construction of Site‐Specific Ground Motion Coherence Functions for Nuclear Power Plant Based on Multi‐Scale Random Sampling Technology

ABSTRACT The spatial variability of ground motion (SVGM) has been shown to induce additional torsion and amplify seismic responses in nuclear structural analyses, necessitating the incorporation of SVGM into design and assessment. However, dense long‐term strong motion arrays are rarely available at specific nuclear power plant (NPP) sites, and coherence functions statistically derived from other arrays lack transferability. Therefore, an engineering approach for site‐specific coherence function is required. This study develops a random field finite element model for a 2D site using a multi‐scale random sampling technique: a coarse‐scale standard normal field is first generated, its statistical characteristics are mapped to fine‐scale elements, and an inverse of the cumulative distribution function (CDF) is applied to achieve the target distributions. Time domain site response analyses are then performed for multiple model samples of a hypothetical site; cross power spectral densities computed from the simulated records are normalized to yield ensemble‐averaged coherence functions. The proposed multi‐scale random field model balances sampling accuracy and efficiency and effectively reproduces heterogeneous site features. For a hypothetical pulse‐type incident wave site, the coherence functions in the horizontal and vertical directions are estimated. The coherence functions for different station spacings show that the coherence decreases with increasing spacing and frequency, with local peaks close to the resonant frequencies. The workflow provides an operational method for constructing site‐specific coherence functions for NPP sites in the absence of array data, laying the technical foundation for incorporating ground‐motion incoherence into seismic analyses of NPP structures at complex sites.