The Characteristics Of Micro-seismic Signal Extraction And Identification Reasarch

Security monitoring microseismic monitoring technology has been widely used in large-scale geotechnical engineering and construction of large structures to run abroad, such as the monitoring of slope stability, the stability of underground caverns and tunnels and other aspects. Although in this technology in the domestic application engineering in the neighborhood just started, but it plays an increasingly important role. Micro-seismic monitoring can provide construction and safety monitoring of large-scale projects based on accurate data, how to quickly and efficiently identify the micro-seismic signals, the safety assessment of major projects have a decisive significance.Micro-seismic monitoring system to monitor because of its high sensitivity,while the micro-seismic signal energy is weak, vulnerable to background noise, these factors have led to the number of signal interference micro seismic monitoring system to monitor very large so that the micro-seismic signals really difficult to be screened micro-seismic signals and extended screening time, reducing the timeliness of the micro-seismic signals, adversely affecting large geotechnical construction and operation of large-scale structure of the safety monitoring.By extracting the micro-seismic signals of different signal characteristics, and produce them after the integration of new features, using a different method of classification and identification of these new features are classified experiment, to find a means to quickly and efficiently identify the micro-seismic signals. In this paper, experimental platform for the Matlab 2013 a, micro-seismic signal feature extraction wavelet packet coefficients are Shannon entropy, wavelet packet coefficients energy ratio, the zero-crossing signal rate, signal kurtosis, signal to noise ratio of the five signal characteristics, classification methods of SVM support vector machine, K-nearest neighbor(KNN) KNN-SVM algorithm and these three algorithms.For more than five kinds of signals generated merging features 12 kinds of signal characteristics, and the use of these three classification algorithm experiment, select the correct category in which a combination of the highest rates of signal characteristics and classification, and then apply it to the actual micro-seismicidentification and classification of signals achieved good results.In this experiment, SVM algorithm and KNN algorithm selected db7, db3, rbio1.5 three different wavelets were tested. First, the choice of SVM support vector machine to classify the extracted feature, in order to test the impact wavelets, classifier, kernel function, different signal characteristics set parameters such as the classification result of each test by changing only one parameter test, in order to observe the effects of various parameters on the result of the classification, SVM classifier algorithm testing selected C-SVC and V-SVC both classifications, a kernel function selected Rbf, Linear,Sigmoid three kinds kernel function results SVM algorithm, using zero-crossing signal, wavelet packet coefficient signal to noise ratio, and kurtosis of the signal, the signal db7 wavelets Shannon entropy combination of features plus Rbf kernel and V classification results-SVC classifier with the best time of 95%. KNN algorithm experiments in selected Kvalue chosen 3,5,7,9,11 these five values, the KNN algorithm results when K = 7, wavelet packet coefficients db7 wavelets Shannon entropy and signal kurtosis, By combination of features of the signal to noise ratio of the best classification accuracy was 85%. In KNN-SVM algorithm Wavelet Bases experiments performed better choice in front of db7 wavelets and V-SVC classifier, K is 2,3,4,5,6, kernel function selected Rbf, Linear, Sigmoid three seed kernels are classified experiment, regardless of when the kernel function K = 5 or 6, characterized by a combination of classificati on accuracy of wavelet packet coefficients db7 wavelets Shannon entropy and signal kurtosis is preferably 55%. Final results showed that V-SVC classification using SVM algorithm, the wavelet packet coefficients zero-crossing signal, signal to noise ratio, signal kurtosis, signal db7 wavelets Shannon entropy signal characteristics combined with Rbf kernel function is to match the actual micro-seismic signal can be effectively screened. This means making the construction and monitoring of personnel to microseismic monitoring system is fast, accurate and reliable extraction of the micro-seismic monitoring signal needs to pass.