Pre-earthquake anomaly extraction from borehole strain data based on machine learning

Abstract

Borehole strain monitoring plays a critical role in earthquake precursor research. With the accumulation of observation data, traditional data processing methods struggle to handle the challenges of big data. This study proposes a segmented variational mode decomposition method and a GRU-LUBE deep learning network based on machine learning theory. The algorithm enhances data correlation during decomposition and effectively predicts borehole strain data changes. We extract pre-earthquake anomalies from four-component borehole strain data of the Guza station for two major earthquakes in Sichuan (Wenchuan and Lushan earthquakes), obtaining more comprehensive anomalies than previous studies. Statistical analysis reveals similar abnormal phenomena in the Guza station’s borehole strain data before both earthquakes, suggesting shared crustal stress accumulation and release patterns. These findings highlight the need for further research to improve earthquake prediction and preparedness through understanding underlying mechanisms.