Human-induced seismicity in underground mining has significant impacts on productivity, safety, and operating costs. Accurate predictors for microseismic event sources are crucial to minimize disasters such as rockbursts. This study develops machine learning algorithms to predict real-time seismic wave velocities in deteriorating underground mines, using data from Nazarbayev University's School of Mining and Geosciences laboratory. Traditional constant velocity models are imprecise, so the study explores several machine learning models. The best-performing model was Gradient Boosted Decision Tree with an MAE of 7.146 m/s. These findings demonstrate that machine learning algorithms can accurately predict seismic wave velocities in underground mining environments.
In the mining industry, rockbursts are frequent and can result in accidents, production loss, and financial losses (Feng et al., 2017 and Liu et al., 2019). Microseismic monitoring systems are being developed to mitigate the effects of rockbursts by accurately locating the origin of microseismic events. The seismic monitoring system needs automated processing for event detection, wave arrival time, and event location. A changing mining environment requires a seismic velocity model with an anisotropic heterogeneous geological model for accurate event source prediction.
Deep learning (DL) is increasingly popular for microseismic event source location, due to its ability to identify patterns without manual feature engineering. This paper optimizes machine learning algorithms for wave velocity prediction in a changing environment using linear regression, Deep Neural Networks (DNN), and Gradient Boosted Decision Trees (GBDT). The methodology, machine learning models, results, and findings are discussed in detail in the following sections.
The technique of learning from data in order to quickly identify which data items are of relevance is known as data exploration. The raw data should be carefully chosen and cleansed in order to ensure credible exploration findings.
A research team at the Nazarbayev University School of Mining and Geosciences laboratory produced the raw data for this investigation. On a rock cube with blocks that had holes of various diameters, backfilled with various cement/sand ratios, and fractured to mimic various mining conditions, the arrival time of the pulse wave generated by one of the sensors as a source was recorded by the other sensors during data collection. The School of Engineering and Digital Sciences at the same university is where the machine learning application research was conducted.
