Research On Early Warning Method Of Coal And Gas Outburst Based On Deep Learning And Multi-Source Information Fusion

Coal and gas outburst,as one of the major dynamic disasters in coal mine,presents an increasingly serious trend,which seriously restricts the safety production of coal mine with the increasing depth of coal mining.Accurate and reliable monitoring and early warning is the premise and key to effectively prevent and control outburst.EMR and AE monitoring technology has been widely used in coal and rock dynamic disaster monitoring and early warning.It is a scientific and technological problem to filter noise information,identify effective signals and early warning accurately in the monitoring process influenced by factors such as the significant uncertainty of the outburst evolution process and the complex interference factors of the working environment.Given the above problems,this thesis proposed a noise filtering method based on the characteristics of multi-source noise signals through experimental methods and field monitoring methods to obtain effective precursor signals and their time-frequency characteristics;A time-frequency feature recognition model based on EMR or AE signal was established and verified.And the outburst early warning model with EMR,AE,and gas concentration signals based on multi-source information fusion was established and applied in the Jinjia coal mine and No.1 Shiping coal mine.The research work and main achievements are as follows:(1)The variation laws of stress and gas pressure,the waveform,time-frequency characteristics,and energy distribution of EMR and AE signals during the evolution of outburst induced by the experiment were analyzed.The results show that there are significant temporal and spatial differences in the changes of coal stress and gas pressure when outburst occurs;Near the location of outburst,the stress and gas pressure decreased sharply;Far from the location of outburst,the stress increases first and then decreases,while the decrease of gas pressure has hysteresis.During the process of tunneling,EMR and AE effective signals fluctuate significantly,and the signal amplitude increases gradually;Before outburst,the amplitude of the effective signal shows a sudden increasing trend;The high amplitude EMR and AE effective signals with continuous growth trend and the proportion of main frequency band of effective signals can be used as the precursor characteristics of outburst;In addition,the waveforms of EMR and AE noise signals are simple and have no significant main frequency band,which are significantly different from precursory characteristic signals in the time and frequency domain.(2)By analyzing the characteristics of EMR and AE intensity signals and multisource noise signals during the process of on-site tunneling,it was found that there are usually two types of coal mine noise,including the type of sudden increasing "sharp pulse" noise(blasting,moving sensor and coal raking by coal raking machine)and the intermittent increasing "n" noise(roof support and regional drilling).A noise filtering method of EMR and AE intensity signals based on SVD-EEMD is proposed,the optimal time scale of SVD noise reduction is determined to be 8 hours,and finally the reconstructed EMR and AE signals after noise reduction are obtained;Then,transformed the reconstructed signals by short-time Fourier transform,and the precursory characteristics of outburst based on two-dimensional time-frequency image are obtained,which provides a basic basis for the recognition of precursory characteristics of outburst risk.(3)Established a precursory feature recognition model of outburst based on VGG-16,and verify the on-site application by EMR and AE signals of the Jinjia coal mine database.The results show that the recognition accuracy of EMR and AE signals precursory features is 98.00% and 98.40%,respectively;Among them,the recognition accuracy of effective precursor signal is 100%,and there is no missing alarm.The model improved the recognition accuracy of outburst precursor characteristics and provided a reliable evidence source for early warning of outburst based on multi-source information fusion.(4)Established a multi-source information fusion early warning model of outburst based on transitive reliability model(TBM),and apply the model to Jinjia coal mine and No.1 Shiping coal mine for on-site application and verification.Among them,EMR,AE,and gas concentration signals were selected as the evidence sources of early warning model.The results showed that the outburst early warning model based on multi-source information fusion can more comprehensively and dynamically reflect the risk of outburst during the process of tunneling.The verification and complementarity between multi-source information ensure that the early warning results are more reliable,and avoid problems such as missing alarm,wrong alarm,or delayed alarm.The model gives an early warning result of the gas gush out event and coal burst event1-3 days in advance in the Jinjia coal mine.The early warning result of K1 value close to the over threshold and gas concentration over threshold 7-12 hours in advance in the No.1 Shiping coal mine,which verified that the model has high reliability and can realize intelligent identification and advanced early warning of outburst risk.The research results of this thesis provide a theoretical basis for outburst early warning and have practical significance for improving the recognition accuracy of precursor characteristics of outburst,improving the advance and reliability of early warning,and enhancing coal mine production safety and risk control.The thesis contains 90 pictures,19 tables and 259 references.