Early warning and detection of rock burst signal in coal mine

ACKNOWLEDGEMENTS: This work was supported by ministry of education industry-university cooperative education project (Grant No.: 231106441092432) and special research project on teaching reform (Grant No.: 30120300100-23-yb-jgkt03)

Abstract

With the increase of mining depth, formation pressure gradually increases, and the risk of underground coal and rock dynamic disaster also increases, which seriously affects the safe and efficient mining of coal mines. Rock burst has become one of the most serious disasters threatening the safety production of coal mines in China. The purpose of this study is to analyze the characteristics of electromagnetic radiation and acoustic emission signals to judge the occurrence time of interference signals and precursor signals and predict the probability of precursor signals. This research is of great significance to improve the accuracy and timeliness of the mine safety warning system, which is helpful to reduce the occurrence of rock burst accidents and related losses. The data pre-processing and visual analysis of the interference signal were carried out to preliminatively observe the characteristics of the data, and then the four time-domain features of the data, namely, mean value, variance, root-mean-square and skewness, were extracted. Then the frequency features of the spectral energy and main spectrum components of the EMR (electromagnetic radiation) and AE (acoustic emission) interference signal data were extracted using the fast Fourier transform algorithm. The data is preprocessed, the outliers are processed and arranged in ascending order of time, and then the visual analysis is performed. The precursory data features are extracted by the method of extracting data features in the first work, and it is found that the precursory signals have the characteristics of excessive regional time and random changes. Then, the sliding window method is used to extract the features, and a window of 25 data points is established to calculate the mean value, variance and other features. Since the original data presents strong randomness and noise, the wavelet transform is first used to reduce the noise of the data, so as to make the change of the data smoother, and then the data after noise reduction is analyzed. Then LSTM time series prediction model is used to predict the data of the next 50 time points, and the sliding window is used to extract the features of the data. Finally, the extracted features are analyzed with decision tree model to identify potential precursor signals, and the probability of precursor features appearing at different time points is calculated.