针对岩体工程中岩体破裂信号与爆破振动信号难以自动区分的问题，提出了一种基于集合经验模态分解（EEMD）关联维数与机器学习相结合的微震信号特征提取和分类方法。利用EEMD将微震信号分解为本征模态函数（IMF）分量，并从得到的IMF分量中筛选出主分量IMF1~IMF4，再通过相空间重构计算出各个主分量的关联维数，最后将所得到的关联维数作为特征向量，使用SVM方法进行微震信号自动识别，并与其他机器学习方法进行对比分析。试验结果表明：该方法对微震信号的自动识别具有较高的准确率，且基于高斯核函数的SVM的识别效果明显优于逻辑回归（LR）和K-近邻算法（KNN）判别法的识别结果，其准确率达到93.7%。

The microseismic monitoring technique is to evaluate the failure and safety of rock mass indirectly by monitoring the vibration signal caused by the rupture inside the rock mass，and it can provide guidance for ground pressure disaster warning and safety production optimization.In order to accurately analyze the behavior of rock rupture，it is necessary to eliminate the interference of non-microscopic signal.At present，microseismic monitoring system can not recognize microseismic signal automatically.The core problem is that the vibration signal is complex，the waveform characteristic is not obvious，the noise is large and multi-seismic superposition occurs.In order to solve the problem that it is difficult to distinguish the rock burst signals and the blasting vibration signals automatically，a method of feature extraction and classification of microseismic signals based on ensemble empirical mode decomposition（EEMD）,correlation dimension and machine learning was pro-posed.Firstly，the microseismic signals was decomposed into Intrinsic Mode Function（IMF） components by EEMD，and the principal components IMF1~IMF4 were selected from the obtained IMF components，the IMF1~IMF4 component was selected as the main component for phase space reconstruction.The delay time and minimum embedding dimension of each component were obtained by autocorrelation function method and Cao algorithm.Then，accoring to the obtained delay time and embedding dimension，the correlation integral curve of IMF1~IMF4 components was obtained by using the G-P algorithm，and the region with the best linearity of the correlation integral curve was found.The integral curve was fitted by least squares，and the resulting linear slope value was taken as the correlation dimension value，and the obtained correlation dimension was taken as the feature vector for microseismic signal recognition..Finally，the SVM method was used to automatically identify the microseismic signals and compare them with other machine learning methods.The experimental results show that the method has a high accuracy for automatic recognition of microseismic signals.The recognition effect of SVM based on Gaussian kernel function is obviously better than the recognition result of Logical Regression（LR） and K-Nearest Neighbor（KNN） discriminant method.The classification accuracy of gaussian kernel function SVM based on EEMD correlation dimension is 93.7%.Based on the analysis，it is found that the recognition effect SVM different kernel functions is different.The recognition effect of Gaussian kernel function SVM is better than that of linear kernel function SVM and Sigmoid kernel function SVM.Therefore，the feature extraction and classification method based on EEMD correlation dimension and SVM provides a feasible new method for mine microseismic signal classification.

http://www.goldsci.ac.cn/article/2020/1005-2518/1005-2518-2020-28-4-585.shtml