Improved mine microseismic event recognition method combining neural network and transfer learning

The data collected by the coal mine monitoring system contains a lot of noise, which is difficult to accurately identify with traditional methods. It can be processed by deep learning methods to mine the deep-level features of the data, making the identification more efficient. Both natural earthquakes and coal mine microseisms belong to the category of vibration. The waveform characteristics, focal mechanism, and monitoring targets are similar. Transfer learning can transfer the characteristics of large sample seismic data set to the model of small sample microseismic data set. Therefore, this paper proposes an improved mine microseismic event recognition method based on transfer learning. First, in view of the low accuracy of microseismic event recognition in the existing deep learning CNN method, a non-recursive SimRank-based model (SimRank CNN) is proposed. Secondly, in view of the problem that the small-sample microseismic data set is not enough to support the training of the neural network model and the recognition accuracy is not high, a source domain feature transfer learning method based on SimCNN is proposed. In order to make the model have a better ability to extract features. Furthermore, a LSTM-based time-series feature transfer learning method is proposed and the Transfer learning SimCNN transfer model (T-SimCNN) is constructed. Finally, the performance of the T-SimCNN transfer learning model has been improved, and the recognition accuracy of microseismic events can reach 95%.