A CNN-GRU rock burst risk analysis model considering micro-seismic precursor characteristics

Abstract

Micro-seismic signals are important for the analysis of rock burst risk in coal mines. However, microseismic events generally cannot determine whether rock bursts have occurred, and the established models rarely consider the characteristics of micro-seismic, resulting in insufficient model performance. In this study, a microseismic dataset with rock burst risk labels is obtained by tracking and calibrating the characteristics of microseismic events in coal mines, and a micro-seismic risk analysis method based on CNN-GRU model is proposed. This method considers the characteristics of micro-seismic precursors, and employs the time, location and energy of micro-seismic signals to establish the characteristic values. The established initial data set is divided into training set, verification set and test set in time scale, and the problem of imbalance between dangerous and non-dangerous samples is dealt with. Finally, the CNN-GRU model is trained by the training set, and the model with the best effect in the verification set is used for testing, which strictly regulates the generalization ability of the model. This method has achieved good results in rock burst risk analysis of micro-seismic monitoring events in a coal mine, which proves that it is reliable to use deep learning method to analyze the rock burst risk on the basis of selecting appropriate analysis features.

Keywords

• micro-seismic
• rock burst hazard
• convolutional neural network（ｃｎｎ）
• recurrent neural network （rnn）
• label dataset