Microseismic B-Value based Time Series Prediction of Rock Bursts

Abstract

Monitoring the microseismic activity in mines helps to understand the interference range and characteristics of the microseismic activity triggered by mining activities, thereby predicting the future trend of rock burst pressure changes based on temporal characteristics. This paper collects microseismic monitoring data and applies the G-R relationship to calculate the microseismic b-value, finding that before the occurrence of rock burst events, the b-value is abnormally low, falling below the average value, which is used as a predictive indicator. Using a deep learning algorithm combining Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Northern Goshawk Optimization (NGO), Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU), and Self-Attention mechanism, a rock burst prediction model is constructed for advanced prediction of the b-value. Compared with different algorithm models, it has a better predictive effect. The model can effectively learn from the characteristic information of the historical microseismic b-value and make predictions. The accuracy of the advanced single-step prediction is quite high without any obvious lag signs, providing a basis for early warning of rock burst events.