Coal is the main energy and industrial raw material in China. In order to prevent the risks of coal mining and ensure the safety and efficiency of coal mining, this paper established random forest model, ADF and MK test model and logistic regression model, and used autocorrelation function algorithm, Fourier transform and sliding window transformation of time series data to analyze and optimize the characteristics of interference signals and precursor characteristics. For the first work, the huge data is preprocessed, tested, and outliers are removed, and the time characteristics are taken to aggregate and classify the data. For work (1.1), data parameters were sorted out, Fourier transform prediction signal model was initially established, and characteristic values were extracted by wavelet transform. Finally, autocorrelation function algorithm was used to optimize and analyze the results. The final results were shown in Table 4 for the data characteristics of electromagnetic radiation (EMR) interference signals and Table 5 for the interference signals of acoustic propagation signals. For work (1.2), a random forest model is established, with sliding window transformation of time series data, window size is specified, boundary processing mode is specified, and label value is defined as interference and normal. The data set is divided into training set and test set in the way of eighty-two allocation. After AE training, there are three stages of training. Finally, random forest algorithm is used to calculate the electromagnetic radiation interference signal interval. For work (2.1), the model and algorithm of work 1 are used to identify the trend characteristics of the data before the occurrence of electromagnetic radiation and acoustic emission signals, and the signal data are judged to have a "slightly rising" trend. The statistical values are compared by KS trend test, and the trend characteristics of normal and precursor signals are calculated by using the autocorrelation function algorithm. A trend feature table is obtained for the precursory feature data of electromagnetic radiation and acoustic propagation signal. For work (2.2), Augmented dickey-Fuller and MK test models of the system were established. For work 3, a logistic regression model is established to predict the probability of precursor feature data appearing at the last moment of multiple time periods. By using maximum likelihood estimation to train model parameters, the characteristics of the last data collection moment of each time period are predicted, and the probability of precursor feature appearing at each moment is output. As shown in Table 14, the probability of precursor features appearing at the time when the data of multi-classification logistic regression is located is obtained. **************** ACKNOWLEDGEMENTS**************** This work is supported by ministry of education industry-university cooperative education project (Grant No.: 231106441092432), the research and practice of integrating "curriculum thought and politics" into the whole process of graduation design of Mechanical engineering major: (Grant. No.: 30120300100-23-yb-jgkt03), research on the integration mechanism of "course-training-competition-creation-production" for innovation and entrepreneurship of mechanical engineering majors in applied local universities (Grant. No.: CXKT202405), Mechanical manufacturing equipment design school-level "gold class" construction project (Grant. No.: 30120324001).
Read full abstract