Real-time Seismic Wave Velocity Prediction for Accurate Source Location in Underground Mines Based on Machine Learning

The objective of this study is to develop machine learning techniques to forecast seismic wave velocities in real-time amidst the constantly changing conditions of underground mining. As human-induced seismicity in underground mining can negatively impact productivity, safety, and operating costs, it is crucial to have an accurate predictor for the source of microseismic events. To simulate the dynamic conditions of underground mining, laboratory experiments using AE systems and concrete blocks mimicking homogeneous rocks were conducted. The concrete blocks were of various sizes with different hole diameters and lengths representing variations in extraction ratio with mine maturity. Stress-induced fractures were produced using a static cracking agent. The blocks were tested with the voids empty and filled with various cement to sand ratios representing backfill in mining. The data collected from AE measurements, such as seismic wave velocity, arrival time, AE hits, and energy, were utilized to train different machine learning models, including linear regression, neural networks, Random Forest Decision Tree, and Gradient Boosted Decision Tree. The Gradient Boosted Decision Tree model demonstrated the highest accuracy with a mean absolute error of 7.15 m/s. The capability to predict seismic wave velocity in real-time would lead to better identification of the location of seismic events, ultimately improving the safety and efficiency of underground mining operations.