Microseismicity-based short-term rockburst prediction using non-linear support vector machine

Microseismic (MS) monitoring is a short-term rockburst prediction technique that foretells the source, time and damage scale inside a rock mass during the rock fracturing process; however, due to the complex underground environment and mechanism of rockburst it is always hard to reliably predict the damage scale (severity) of rockburst manually; therefore, this paper introduces machine learning (ML) approach using nonlinear support vector machine (Nonlinear-SVM) to predict the short-term rockburst. Six indicators, cumulative number of events (N), cumulative seismic energy (E), cumulative apparent volume(V), event rate (NR), seismic energy rate (ER) and apparent volume rate (VR), are selected as an input indices for Nonlinear-SVM which is trained and tested with randomly selected 85 and 22 samples of rockburst cases, respectively, collected from different literature. The constructed model was employed to predict the short-term rockburst severity. After data standardisation, cross-validation and hyperparameter optimisation, the prediction accuracy reached 86% for the test sample. The predicted rockburst result truly matches the actual situation with few misclassifications. Therefore, the proposed method has potential value for the short-term rockburst prediction task.

Automated summary

This paper introduces a machine learning approach using nonlinear support vector machine (Nonlinear-SVM) to predict short-term rockburst. Six indicators are selected as input indices for the model, which is trained and tested with rockburst cases collected from literature. After data standardisation, cross-validation and hyperparameter optimisation, the prediction accuracy reached 86% for the test sample, with few misclassifications. The proposed method shows potential value for short-term rockburst prediction.