Time series prediction model of microseismicity in coal mine based on adaptive ensemble learning

In time series prediction of microseismicity in coal mines, data-driven deep learning model can effectively extract key features and rules of microseismic data, and achieve accurate prediction. However, due to the constant change of working conditions with the advance of working face, the causes of microseismic events are complex and disordered. It is difficult for a single prediction model to learn local time series features to maintain a stable and accurate prediction, and it is impossible to predict the initial stage of mining. To improve the stability, completeness and accuracy of time series prediction of microseismicity in coal mine, the time series characteristics of microseismicity of different working faces and different mining stages of the same working face were analyzed by using massive microseismic data. Based on adaptive-dickey-fuller(ADF) stationarity and volatility tests, the significant influence of event causes on microseismic timing characteristics was identified, and multiple microseismic timing data sets were constructed under different event causes. Using blending integration learning algorithm and combining one-dimensional convolutional neural network (1D-CNN) and bidi-rectional short- and long-term memory network (BiLSTM), a timing prediction model of coal mine microseismicity based on adaptive ensemble learning was proposed. Based on the microseismic data of 7302 working face in Zhaolou Coal Mine, the daily maximum energy, average energy and frequency of microseismic are predicted by the integrated model, and the daily maximum energy is analyzed in detail. The results show that on the premise of ensuring the integrity of prediction, the proposed integrated model can better adapt to the actual situation of complex, variable and disordered distribution of event causes. The error between the prediction and field monitoring values is small, and the goodness of fit calculation results of each parameter is above 0.8. The research provides a new idea for the time series prediction of microseismicity in coal mines.