Deep learning based reduced order modeling of seismogram-type acceleration time series model: Part - I

Deep learning based reduced order modeling

Abstract

The critical slip distance in rate and state model for fault friction in the study of earthquakes can vary wildly from micrometers to few meters depending on the length scale of the critically stressed fault. This makes it incredibly important to construct a Bayesian inversion framework that provides good estimates of the critical slip distance based on the observed acceleration at the seismogram. That estimate eventually helps in predicting future seismic behavior. To eventually construct such a framework for real data, we first work on synthetic data. This data is generated by adding noise to the acceleration output of spring-slider-damper idealization of the rate and state model as the forward model. Furthermore, the forward model in real-time field studies can be incredibly complicated, and constructing a reduced order model is of paramount importance for the Bayesian framework to be a starter in the first place. With that in mind, we use deep learning architecture of LSTM encoder-decoder towards constructing a reduced order model for the spring-slider-damper idealization. This is part-I in a series of research studies.