Abstract:
Microseismic monitoring data plays a crucial role in predicting and warning dynamic hazards such as coal mine rockburst. Utilizing historical microseismic monitoring data to forecast the evolution characteristics of future microseismic events is an effective approach to enhance the timeliness and accuracy of rockburst prediction and early warning. However, due to the non-stationary nature of microseismic data, conventional time series prediction methods often struggle to achieve accurate forecasts. To address this issue, this paper proposes a microseismic time series prediction method based on the combination of modal decomposition technique and deep learning methods. Firstly, the adaptive noise complete ensemble empirical mode decomposition (CEEMDAN) is employed to decompose the microseismic data into multiple intrinsic mode sequences. Subsequently, the decomposed sequences are reconstructed into high-frequency, low-frequency, and trend sequences using sample entropy. Furthermore, the variable mode decomposition (VMD) method is utilized to further decompose the high-frequency and low-frequency sequences into multiple new intrinsic mode sequences, which serve as the multivariate feature time series of microseismic data. Finally, a convolutional neural network (CNN) and long short-term memory network (LSTM) combined microseismic time series prediction model is established, with the microseismic multivariate feature time series within a certain time window and historical microseismic data as inputs, and the next moment microseismic monitoring data as output. The Bayesian optimization method is employed to optimize the hyperparameters of the model, aiming to improve the accuracy of microseismic prediction. Based on the microseismic monitoring data from the multiple working faces of the Xinjulong coal mine, the proposed method is applied to predict the daily maximum energy, daily average energy, and daily frequency of microseismic events, and a series of comparative experiments are conducted on the prediction of daily maximum energy data. The results show that the proposed method can effectively predict the evolution trend of microseismic events, with small errors between the predicted results and actual monitoring values. It demonstrates a good predictive performance and generalization ability.