Machine learning-based on-site Earthquake Early Warning

Abstract

Worden et al., 2012). However, the prediction of PGV or PGA and GMICE were used simple regression techniques with limited data, and errors are accumulated through two empirical equations to calculate seismic intensity. Therefore, in this thesis, a machine learning (ML) model that can directly estimate the seismic intensity scale has been developed from initial P-waveforms of three-component acceleration data measured at a single station. 1D-Convolutional Neural Networks (1D-CNN) was used as a ML model, and K-Net and KiK-net datasets were used as training data. The seismic intensities were reasonably estimated with a mean absolute error (MAE) of 0.33 and a coefficient of determination of 0.69. The 97% of estimated data showed the prediction values within ±1 difference in its true seismic intensity (IJMA). Transfer learning (TL) technique considering imbalance problem between datasets used in pretraining and new target dataset were adopted to apply the ML model pretrained with a specific dataset to a new dataset obtained from other regions. STEAD dataset (Mousavi et al., 2019a) was used as a new target data for other regions with imbalance problem. When applying TL, the consideration of imbalance problem in training data showed improvement of prediction results especially for the target data which were not sufficiently used in pretraining. The seismic intensity estimation results of applying TL to STEAD data showed that the ML model reasonably fits the target data with a coefficient of determination of 0.75. Therefore, it was confirmed that the developed 1D-CNN on-site EEW model trained with K-Net and KiK-net datasets can be successfully modified to predict seismic intensities of a new dataset acquired from other regions using TL considering imbalance problem.; Earthquake Early Warning System (EEW) is a technology that calculates earthquake parameters using P-wave that arrives earlier, and warns the expected damage area before the arrival of destructive S wave. Therefore, many countries are operating EEW to mitigate damage from earthquake shaking. Especially an on-site EEW is drawn attention as it can reduce blind zones due to using only a single or minimum station. The conventional studies for on-site EEW calculate seismic intensity scale from predicted Peak Ground Velocity (PGV) or Peak Ground Acceleration (PGA) through Ground Motion Intensity Conversion Equation (GMICE) (Colombelli et al., 2015