Key points are not available for this paper at this time.
Abstract Ground motion models (GMMs) are a fundamental component of probabilistic seismic hazard analysis (PSHA) and play a critical role in guiding structural seismic design by predicting ground motion parameters under specified earthquake scenarios. Since most structures enter a nonlinear state under the excitation of strong ground motions in big earthquake, the use of inelastic response spectra (IRS) for PSHA, structural seismic design and performance evaluation is considered more appropriate. In this study, the mixed-effects constant-strength inelastic spectral acceleration (ISA) prediction model was established by using 15,512 records from 171 earthquake events in the NGA-West2 database based on the XGBoost algorithm. The input parameters included moment magnitude (M w), Joyner-Boore distance (R JB), source parameters (Strike, Tplunge, Pplunge, Dip, and Rake angle), average shear-wave velocity up to 30 m depth (V S30), depth to the top of the rupture (Z tor), depth to the sedimentary layer with shear-wave velocity of 1 km/s (Z 1), fault type (F t), as well as structural parameters (period T and strength reduction factor R y). To ensure the robustness of the proposed model, both within-event and between-event residual were analyzed to confirm its reliability and unbiasedness. Finally, the feature interpretability analysis was performed using permutation importance and SHAP methods, and the prediction accuracy was compared with the results of traditional GMM equation. The results demonstrate that the proposed machine learning model achieves high performance in predicting ISA, with period T, magnitude M w, and distance R JB identified as the most influential features.
Gong et al. (Thu,) studied this question.