Prediction models of Newmark sliding displacement considering pulse-like ground motions

Abstract

Several machine learning-based Newmark sliding displacement prediction models have been developed by researchers to assess the seismic performance of numerous slopes in a region. Sliding displacements induced by pulse-like ground motions (PGMs) are larger and cause more severe damage. However, existing machine learning-based models cannot accurately predict PGMs-induced sliding displacements. In this research, a Newmark sliding displacement prediction model considering PGMs is developed by improving in two aspects: sliding displacement generation and intensity measurements (IMs) selection. The improvement in sliding displacement generation can avoid unfavorable underestimation of sliding displacements. While the improvement in IMs selection can increase the efficiency of models in previous studies, the R² is improved by up to 83.71% and the RMSE is reduced by up to 45.49%. In addition, the proposed prediction models can satisfy the sufficiency requirement.