Prediction Model for Peak Ground Acceleration Using Deep Learning

Over the last decade, several studies have been proposed in the field of earthquake early warning (EEW) systems. Deep learning can be used to determine the magnitude of earthquakes and predict the PGA (peak ground acceleration). Earthquake catalogs are essential for studying fault systems, modeling seismic events, assessing seismic hazards, predicting them, and eventually decreasing seismic risk. In this work, the seismic hazard analysis is given along with the scale of ground vibration in terms of peak ground acceleration (PGA), which would be crucial for constructing earthquake-resistant structures, i.e., the PGA earthquake prediction is crucial. We propose to use artificial neural networks (ANN) and convolutional neural networks (CNN) to predict the PGA using the waveforms of weak motion velocity recorded in Japan. In this study, we use 555 events recorded by 5 seismic stations (velocity data) where the magnitude (Mg) is larger than 3. The selected earthquakes occurred between 2003 and 2022 recorded by the K-NET, Kiki-NET, and Hi-Net networks. As a result, the mean absolute error (MAE) for the test set is 18.23.