Automation of seismic events classification during seismic monitoring at a coal mine using machine learning

Abstract

The paper is dedicated to the development of an algorithm for automatic classification of geodynamic processes in the context of monitoring seismic activity in mines using machine learning methods. The importance of classifying geodynamic processes is noted in terms of understanding the nature of seismic phenomena, identifying their sources, assessing the potential hazard as well as their impact on the environment and the infrastructure of underground structures. The paper describes an algorithm for analyzing seismic activity based on the data obtained as the result of recording seismic events using the hardware complex and the GITS2 seismic monitoring software in a coal mine. The paper briefly examines the key artificial intelligence methods used to control and predict hazardous geodynamic phenomena. Particular attention is paid to the development of a machine learning model based on decision trees that demonstrates high accuracy in classifying seismic events. The classification accuracy of the developed model is 98,39% on the training set and 98,41% on the test set. This result indicates the high generalization ability of the model on new data and the absence of overfitting. Testing the algorithm on new data entering the system also confirms the high accuracy in classification of seismic event types with the level of 83–93%. This highlights the efficiency of machine learning methods in mine seismic control. After its trial operation, the developed machine learning model will be implemented in the GITS2 monitoring system, that will allow classification of the incoming seismic events in automatic mode.