1. Key Laboratory of Mining Engineering of Heilongjiang Province College,Heilongjiang University of Science & Technology,Harbin　 150022,China; 2. Col-lege of Mining Engineering,Heilongjiang University of Science & Technology,Harbin　 150022,China

针对岩体破裂信号与爆破振动信号难以自动识别的问题,提出了基于局部均值分解(LMD)和模式识别的矿山微震信号特征提取及分类方法。首先采用LMD对微震信号进行自适应分解得到乘积函数(PF)分量,再利用相关系数和方差贡献率筛选得到PF主分量,进而计算各主分量的相关系数和能谱系数,并以此作为模式识别的特征向量。结果表明:LMD、经验模态分解(EMD)和离散小波变化(DWT)的主分量分别为PF1~PF6,IMF1~IMF6和D2~D7,其中IMFi(i=1,2,…,6)为EMD分解的本征模态分量,Dj(j=2,3,…,7)为DWT分解的细节分量;LMD主分量分类识别结果整体上优于EMD和DWT主分量分类识别结果;能谱系数分类结果整体上优于相关系数分类结果,人工神经网络(ANN)和支持向量机(SVM)识别效果明显优于逻辑回归(LR)和Bayes判别法识别结果,且基于LMD能谱系数的SVM分类准确率达到了93.0%。

To solve the problem that it is hard to classify rock mass fracturing signal and blasting vibration signal auto- matically,a feature extraction and classification method of mine microseismic signals based on local mean decomposi- tion (LMD) and pattern recognition was proposed. Firstly,the LMD was used to decompose a microseismic signal a- daptively and obtain PF (product function) components,then correlation coefficients and variance contribution ratios were applied to select PF main components. Finally,the correlation coefficients and energy spectrum coefficients of the selected main components were set as the input of pattern recognition. The results show that the main components of LMD,empirical mode decomposition (EMD) and discrete wavelet transform (DWT) are PF1 to PF6 ,IMF1 to IMF6 and D2 to D7 respectively,where IMFi( i = 1,2,…,6) means the intrinsic mode function of EMD and Dj( j = 2,3,…,7) corresponds to the detail component of DWT. In addition,the classification result based on LMD is better than that based on EMD and DWT. Furthermore,the energy spectrum coefficients based method obtains a better result than that of correlation coefficients based method,and the artificial neural network (ANN) and support vector machine (SVM) based methods are much better than that of logistic regression (LR) and Bayes based methods. Also,the classification result based on energy spectrum coefficients of LMD and SVM obtains the best accurate classification rate of 93. 0% .

黑龙江省普通高等学校采矿工程重点实验室开放课题资助项目(2014KF04)；黑龙江省自然科学基金面上资助项目(E2016061)；

Li Wei. Feature extraction and classification method of mine microseismic signals based on LMD and pattern recognition[ J]. Journal of China Coal Society,2017,42(5):1156-1164.