Rock Burst Precursor Electromagnetic Radiation Signal Recognition Method and Early Warning Application Based on Recurrent Neural Networks

Abstract

The fracture-induced electromagnetic radiation (FEMR) method is a promising geophysical method for the monitoring and early warning of coal-rock burst disasters. At present, the time-series characteristics of FEMR data are primarily used for rock burst monitoring and early warning. However, these rock burst precursor signal identification and hazard warning methods need further development. The progress made in the field of deep learning provides a new method for the identification of rock burst warning signals and the realization of intelligent early warning. In this paper, based on a deep learning algorithm, a method for identifying rock burst precursor FEMR signals is proposed. Based on bidirectional long short-term memory recurrent neural networks, this method trains and validates the model by analyzing a large number of normal FEMR signals and rock burst FEMR precursor signals. The input of the model is the FEMR data sequence, and the corresponding output is the hazard identification result. Upon the completion of training and validation, the model can perform automatic/intelligent shock hazard precursor signal recognition and quickly and accurately provide rock burst hazard early warning without requiring parameter adjustment and manual intervention. The results obtained showed that the rock burst precursor signal recognition method based on the recurrent neural network responds well to the rock burst hazard and can capture information regarding impact hazards in advance. Therefore, it is of great significance with regard to accurate rock burst monitoring and early warning.