| Rock burst is a kind of dynamic phenomenon, which is mainly caused by thestress of coal and rock mass around working face or ming roadway exceeds thelimiting value, and the elastic energy in them is suddenly released. With the increasingof mining depth and mining scale year by year in China, rock burst is becoming oneof the main disaster in mine especially in deep mining, and seriously threatens to thesafe and efficient production of coal mine.Electromagnetic radiation (EMR) technology is a promising non-contacttechnology for forecasting rock burst disaster, and it has been already used to monitorrock burst in mine. There are many types electrical of equipment in mine, theygenerate massive amounts of electromagnetic disturbance, and the EMR of coal orrock is very weak. The interference in the narrow space of mine has serious effects onthe EMR monitoring or acquisition. At present, there are three key problems to besolved: one is to get the EMR information completely and continuously, another is torecognize the electromagnetic interference signal and eliminate it, and the other is toconstruct a model for extracting omen information of disaster.To meet the research and application needs of the EMR technology, thisdissertation makes systematic study on the recognition of the portentous informationof rock burst. The main research contents are as follows:(1) The continuous and real-time monitoring methods of EMR underground thecoal mine are studied and a new monitoring system is developed. The system canmonitor multi areas, and capture the signal wave speedily and continuously, and becompatible with the common communication network, such as RS232, RS485, CANbus, etc... The host on ground can get the data real-time and change the workingparameters by teleoperation.(2) The electromagnetic interference recognition is studied. In order to avoidomitting effective recognizable features, they are extracted from time domain, spectraldomain and wavelet domain, and compose a high-dimension feature vectors. Onaccount of “curse of dimensionality†and the Lknorm distance measure decrease inresolution ability, a novel general distance, Dffusing distance, and a feature selectionmethod, Df–Relief algorithm are proposed. Experiments on ten UCI classification data sets, which are completed by eight classification algorithms, show that Df–Reliefalgorithm is effective and robust. The algorithm is employed for the feature selectionfrom the high-dimension feature vectors sets, and the naive Bayes classifier and otherscan recognize the electromagnetic interference by the selected feature vectors sets.(3) The nonlinear filter method EEMD-ACMF (ensemble empirical modedecomposition adaptive morphological filter) is proposed. The numerical experimentresults confirm that the filter restrains the random noise and white Gaussian noisemixed in simulated signal effectively. It is shown that the signal-to-noise ratio of EMR,which is captured in mine, filtered by EEMD-ACMF is improved obviously.(4) In order to selected appropriate model for extract omen information fromEMR time series, the series’ non-Gaussianity, nonstationarity and non-linearity aretested respectively by Kolmogorov-Smirnov test, adf/pp test and Hinich test.(5) The omen information recognition model based on SVM is established. Dueto the serious insufficiency of negative-class data in rock burst monitor system, theestablishment (training) of the model is typical small-sample machine learningproblem. According to that SVM is based on statistical learning theory, it can solvethe problem better. The model is verified by true data captured in mine, and theexperiment result indicates that the model has good generalization capability. |
PDF Full Text Request