Current Issue

EARTHQUAKE MONITORING USING DEEP LEARNING WITH CNN–TRANSFORMER ARCHITECTURES

Yu-Chien Wu*¹

*¹Ph.D., Ph.D. Program for Infrastructure Planning and Engineering, 

Feng-Chia University, Taiwan. Email:This email address is being protected from spambots. You need JavaScript enabled to view it.

Tse-Shan Hsu

President, Institute of Mitigation for Earthquake Shear Banding Disasters

Professor, Feng-Chia University, Taiwan.

Zong-Lin Wu*², Yan-Ming Wang*¹

*²Assistant Professor, National Chin-Yi University of Technology, Taiwan.

Tsai-Fu Chuang, Da-Jie Lin

Associate Professor, Feng-Chia University, Taiwan.

Abstract

This study adopts the Chelungpu Fault and the disaster-affected Zhongxingxin Village in Taichung, Taiwan as case studies. We propose an earthquake monitoring framework based on a deep learning architecture. Strong-motion records are used to construct a near-fault training dataset and a nationwide testing dataset, enabling a comparative evaluation of four models, including a CNN–Transformer model. For near-fault earthquakes ranging from minor to moderate magnitudes, the results demonstrate that the CNN–Transformer model exhibits superior responsiveness and higher F1-scores in classification and identification tasks. Moreover, it shows the smallest performance degradation under cross-regional testing. The attention mechanism further enhances interpretability by effectively capturing the initial P-wave amplitudes and the characteristic energy frequency bands of S-waves. Consequently, under conditions where earthquake early warning is available, the proposed approach can contribute to earthquake disaster mitigation.

Keywords: earthquake, early warning, disaster mitigation, fault, CNN–transformer