Probabilistic Three-Component Ground Motion Time History Generation Modeling Using Deep Generative Model

Abstract

Towards probabilistic seismic hazard assessment based on ground motion time history data, we develop a probabilistic model capable of directly generating three-component acceleration time history data of ground motion. Recently, generative models using deep learning have attracted significant attention due to their high performance. In this paper, we employed a deep generative model called Generative Adversarial Networks to learn from the strong-motion records of crustal earthquakes. The model after training is capable of generating ground motion time history data consistent with the conditions of magnitude and distance, and is a probabilistic model approximating the distribution of learned database. Furthermore, it is shown that the distribution of the generated ground motion generally corresponds to the existing ground motion prediction equations.