Research On Early Warning Of Rock Burst Microseismic,Acoustic Emission And Electromagnetic Radiation Signals Based On Deep Learning

As coal resources enter the stage of deep mining,the danger of rock burst disasters is increasing,which seriously affects the efficient production of mines and personnel safety.Accurate and reliable monitoring and early warning of rock burst disasters can effectively reduce the occurrence of rock burst accidents and reduce the risk of rock burst.Microseismic,acoustic emission and electromagnetic radiation monitoring methods are widely used in the field of monitoring and early warning of rock burst disasters.At present,the time series characteristics of monitoring data are mainly used for monitoring and early warning of rockburst hazards,and the identification of precursor signals and danger early warning of rockburst hazards lack intelligent means.The development of deep learning provides a new method for the identification and intelligent early warning of rockburst hazard precursor signals.In this thesis,combined with on-site monitoring and deep learning methods,a deep learning-based microseismic waveform recognition model and acousto-electric interference signal recognition model are built,and the interference identification and filtering of the original microseismic,acoustic emission and electromagnetic radiation signals are carried out.A deep learning-based microseismic,acoustic emission and electromagnetic radiation precursor signal recognition model was built to identify the microseismic,acoustic emission and electromagnetic radiation precursor signals,and the accuracy of the precursor signal identification results was improved;a deep-learning-based microseismic,acoustic emission and electromagnetic radiation signals prediction model was built,compared and analyzed the prediction results of different models;finally,established a comprehensive early warning method of rockburst microseismic,acoustic emission and electromagnetic radiation signals based on deep learning,studied and predicted the changes in the degree of danger during the formation of rockburst disasters,and carried out combined with actual cases.Research on the early warning effect of rock burst danger.The research work and the main results obtained are as follows:(1)A classification and recognition model of microseismic waveforms based on Res Net-50 convolutional neural network and a classification and recognition model of acoustic and electrical interference signals based on cyclic neural network are established,and the performance of the model in the test set is analyzed.An explanation is given,and a generalization analysis of the model is carried out using the microseismic waveform data and the acousto-electric interference data of different time periods and regions.The results show that the comprehensive recognition accuracy of the convolutional neural network model to the test set microseismic waveform signals is98.6%.The interference classification kappa coefficient of the cyclic neural network model for the acoustic emission test set is 0.9926,and the interference classification kappa coefficient of the electromagnetic radiation test set is 0.992,and the classification and recognition accuracy is high.The convolutional neural network model has a comprehensive accuracy rate of 98.4%for the recognition of coal-rock fracture microseismic waveform signals in different regions and time periods.The kappa coefficients of the cyclic neural network model for the recognition of acoustic and electrical interference signals are 0.983 and 0.977,respectively.The model has a generalization ability.(2)A classification and recognition model of rockburst and microseismic,acoustic emission and electromagnetic radiation precursor signals based on cyclic neural network was established,and the identification results of precursory signals were analyzed.The classification and recognition errors of the model are analyzed,and the superiority of the model is proved.And the generalization analysis of the model is carried out.The results show that the model has a comprehensive accuracy rate of 89.05%for the classification and recognition of precursor signals in the acoustic emission test set,87.8%for the classification and recognition of precursor signals in the electromagnetic radiation test set,and 92.93%for the classification and recognition of precursor signals in the microseismic test set,the classification and recognition accuracy is high.The classification effect of the model on the microseismic,acoustic emission and electromagnetic radiation precursor signals of different working faces is still very good.The comprehensive accuracy rates of the recognition of the microseismic,acoustic emission and electromagnetic radiation precursory signals of the shock ground pressure are 96.7%,94.7%,and 93.9%,respectively.The model has a good generalization accuracy.(3)Continuous Fourier transform is performed on the data after the classification of microseismic waveforms and the processing of acoustic and electrical interference,and the long-term and short-term development trends of the microseismic,acoustic emission and electromagnetic radiation signals are captured.A variety of microseismic,acoustic emission and electromagnetic radiation signals prediction models(single-input,multi-input time series prediction models)based on deep learning are established.By plotting the predicted value and the real value of the microseismic,acoustic emission and electromagnetic radiation signals and calculating the evaluation indicators such as MAE,RMSE,MAPE,R~2,the prediction results of the single-input ARIMA,CNN,and RNN models in the microseismic,acoustic emission and electromagnetic radiation signals prediction task were qualitatively and quantitatively analyzed,and quantitatively analyze the prediction results of the multi-input CNN model.The results show that the single-input CNN and RNN prediction models can make separate predictions of the microseismic,acoustic emission and electromagnetic radiation signals.The multi-input CNN prediction model can carry out comprehensive prediction of the microseismic,acoustic emission and electromagnetic radiation signals,and the prediction speed is faster than that of the single-input prediction method.The R~2 values of CNN and RNN models are both above 0.7,and the prediction effect is good.(4)Established a rockburst risk analysis model based on Python data analysis and multi-input convolutional neural network,studied the risk degree and risk change trend of the whole process of rockburst,and predicted future rockburst through current signals of danger.Combined with the classification and identification model of disturbance signal,the identification model of rockburst precursor and the prediction model of rockburst danger,a comprehensive early warning model of rock burst is built,which improves the accuracy and timeliness of early warning of rock burst disasters.The results show that the microseismic,acoustic emission and electromagnetic radiation comprehensive early warning model of rockburst can comprehensively analyze the precursor information and prediction information of the three kinds of signals to obtain the rockburst risk coefficient,and eliminate the microseismic,acoustic emission and electromagnetic radiation monitoring method for the precursory response of rockburst disasters.The microseismic,acoustic emission and electromagnetic radiation signals comprehensive early warning model can issue a rock burst hazard warning 1 to 3 days before the occurrence of rock burst,which is more accurate and time-sensitive than the traditional method.The research results of this thesis provide an intelligent comprehensive early warning method for the risk early warning of rock burst disasters,which can improve the recognition accuracy of rock burst precursor signals,improve the timeliness and reliability of rock burst early warning,and ensure efficient mine production and personnel safety has real meaning.The thesis has 84 figures,38 tables,and 171 references.