Neural-network-based models for generating artificial earthquakes and response spectra

The purpose of this paper is to develop efficient neural-network-based models for the generation of artificial earthquake and response spectra. Five neural-network-based models are proposed for replacing traditional processes. Artificial-neural-network-based model-I (ANN-I) substitutes the parameter identification process of empirical model for generating Fourier amplitude spectrum. ANN-II and ANN-III are trained to obtain the parameters of the power spectral density function and the intensity function, respectively. ANN-IV directly generates an acceleration response spectrum with basic information such as magnitude, epicentral distance, site conditions and focal depth. Finally, ANN-V inverses the ANN-IV, so that it can be applied to the generation of synthetic ground motion accelerograms compatible with a target response spectrum. This study shows that the procedure using neural-network-based models is applicable to generate artificial earthquakes and response spectra. Several numerical examples are given to verify the developed models.