A multivariate time series prediction model for microseismic characteristic data in coal mines

Abstract

Rock burst disasters in coal mines have become a growing concern, posing significant risks to operational safety. Utilizing historical microseismic data to predict future microseismic events can provide effective prediction and early warning for rock bursts. This study proposes a multivariate microseismic sensitive features prediction network model named Deformer, which can accurately predict multiple sensitive feature values extracted from microseismic monitoring data and provide data support for the early warning and prevention of rock bursts. Deformer integrates Transformer and signal decomposition methods, considering both feature and temporal correlations. It enables a comprehensive and in-depth analysis of the relationships among multi-dimensional sensitive features and the temporal evolution of each feature. We extract three characteristic values from the microseismic monitoring data of a coal mine in Shandong Province: daily total energy, daily maximum energy, and daily average energy, and predict the daily maximum energy. By comparing with various classical time series prediction models, Deformer achieved the best results in mean square error (MSE), mean absolute error (MAE), the coefficient of determination (R²), root mean square error (RMSE), and Theil's inequality coefficient (TIC), proving Deformer's significant advantage in predicting microseismic sensitive features. Additionally, testing on various public datasets, such as those for electricity and weather, further validates the model's generalization capability.