2025 Volume 81 Issue 13 Article ID: 24-13498
The use of response spectra is an effective engineering technique to account for the periodic characteristics of seismic ground motions. With the progress of research on the characteristics of seismic ground motions, ground motion models have been developed into more complex functions. On the other hand, non-parametric ground motion models, such as machine learning, have been proposed. The advantage of non-parametric models is that it is not necessary to consider the functional form in advance. This study quantitatively investigated the applicability of Gaussian process regression, one of the non-parametric methods, to acceleration response spectra. Gaussian process regression has the problem of steeply increasing computational cost as the amount of learning data increases. Therefore, by applying the sparse approximation to large amounts of learning data, good results were obtained while reducing the computational cost. In order to obtain the best possible ground motion model, several issues in the Gaussian process regression were investigated.
