Nonlinear effect assessment for seismic ground motions of sedimentary basins based on deep neural networks

Rapid post-earthquake assessment of nonlinear features in geotechnical soils within sedimentary basin is crucial for quantifying site response and seismic risk zoning. However, traditional methods like the classical spectral ratio approach suffer from drawbacks such as insufficient effective data and low efficiency in calculating nonlinear degree indexes for evaluating nonlinear features. To address this issue, this study explores the use of deep neural network (DNN) algorithms as a solution. Initially, sites within sedimentary basin in Japan are identified. The results of horizontal-vertical spectral ratios (HVSR) and different proxy conditions (ground motion intensity and site conditions) are utilized to develop and train DNN models. The dependence of the nonlinear features on various combinations of ground motion intensity and site conditions is analyzed by the DNN model. Based on the differences between the values obtained under weak and strong earthquakes, evaluation indexes of nonlinear features, including the degree of nonlinearity (DNL), absolute degree of nonlinearity (ADNL), and percent nonlinear site response (PNL), are calculated. This allows a rapid assessment of the regional nonlinear features of sedimentary basins. The DNN model is used to determine the nonlinear features of several soil profiles under different ground motion intensity conditions. The results demonstrate a strong consistency between DNL, ADNL, and PNL with variations in ground motion intensity, while showing weaker consistency with site conditions. Finally, a real earthquake case study is incorporated to assess the practicality of the proposed procedure. This study provides a reference for the study of earthquake engineering problems using DNN models.