Machine learning-based prediction of site responses at liquefiable sites subjected to bi-directional ground motions

抄録

In reality, when seismic waves arrive, sites are shaken by multi-directional ground motions that are usually decomposed into three components: two perpendicular horizontal (E-W and N-S) and one vertical (UP). Numerous experimental and numerical studies have demonstrated that two horizontal (or bi-directional) motions can significantly aggravate the building and site responses compared to the scenario where only one horizontal motion is applied. However, 1D site response analysis (SRA) is still prevalent in quantifying site effects and responses due to its simplicity. Recently, we investigated the effect of bi-directional ground motions on site responses at potentially liquefiable sites. Results proved that bi-directional ground motions can intensify the site responses and increase the possibility of getting liquefied. In this study, we utilize a machine learning-based method to predict the soil liquefaction under four different excitation conditions, including 1D SRA using E-W, N-S and RotD100 motions, and 2D SRA subjected to bi-directional motions. Moreover, we evaluate the relative importance and correlations of a broad range of ground motion and soil parameters, i.e., peak ground acceleration (PGA), peak ground velocity (PGV), spectral acceleration, relative density, and V_{s, 30}, in evaluating the liquefaction hazards.