Deep transfer learning for P-wave arrival identification and automatic seismic source location in underground mines

Seismic monitoring routines provide a robust framework for assessing rock stability and dynamic hazards in underground mining operations. However, the labor-intensive task of manually identifying wave arrivals and the suboptimal selection of geophone arrays do not meet the stringent timeliness and accuracy necessary for seismic source location in such contexts. The precise identification of wave arrivals in mining-induced seismicity and the automated selection of an optimal geophone array have emerged as critical challenges in achieving high-performance seismic monitoring in underground mines. To address these challenges, this paper introduces a novel deep transfer learning approach for identifying seismic wave arrivals, and developing an automatic geophone array selection method for seismic source localization in underground mines. First, an initial deep-learning model was constructed using a substantial seismic dataset comprising global earthquakes, designed to detect the arrival of seismic waves automatically. Then, a deep transfer learning process was applied, leveraging a seismic dataset of over 8,000 carefully picked P-wave arrivals from mining environments. This additional training enabled the model to adapt to the unique characteristics of mining-induced seismicity. In parallel, we introduced an innovative method to select geophone arrays based on mine-planned blasting sources. This approach determines the geophone array that minimizes location errors while reducing the standard deviation of P-wave arrivals compared to historical blasting sources. The effectiveness of this method was validated using recorded blasting data from a longwall panel in an underground coal mine. The results demonstrated a median horizontal locating error of 48.95 m, which can be further minimized to a range of 0 m to 17.63 m when considering systematic biases in seismic monitoring. These findings confirm the practicality and feasibility of our method, offering a valuable solution for the automation and enhancement of high-precision seismic monitoring in underground mining operations.