基于小波分形特征与模式识别的矿山微震波形识别研究

在前人研究的基础之上,探讨,验证了矿山微震信号的分形特征,并根据微震信号的特征确立了相关的无标度区间及分形盒维数算法.利用矿山爆破振动,岩石破裂及电磁干扰3类信号频谱分布不同这一特征,结合小波分析与分形理论,通过MATLAB编制相应程序,对3类信号进行了5层小波包分解,求取了指定特征频带上重构信号的分形盒维数.以该盒维数为信号特征,建立23维特征向量,通过SVM支持向量机对300组矿山现场微震信号进行了训练和分类识别.研究结果表明特征频带上的分形盒维数与信号的整体盒维数接近,表明特征频带上的分形盒维数表征了整体的分形特征;3类信号具有明显分形特征,电磁干扰信号最为明显;SVM识别网络识别微震信号,识别正确率高,但运算速度及识别效率仍有待提高.

Classification of mine microseismic events based on wavelet-fractal method and pattern recognition

High-precision location is the focus of hot research about microseismic technology, and classification of microseismic events is the basis for the location. Based on the previous studies, the fractal characteristics of microseismic signals are studied, and the range and scale-free fractal box dimensions of the algorithm are established. Using the spectral differences about mine blasting vibration, rock fracture and electromagnetic interference signals, and based on the wavelet analysis and the fractal theory, the microseismic signals are decomposed into 5 layers to gain specified frequency bands using MATLAB software. Then the box fractal dimensions about those specified frequency band reconstructed signals can be calculated. The 23-dimensional values of pattern recognition feature vector can be established. Finally, the support vector machine SVM is adopted to train, classify and recognize 300 sets of data. The results show that three types of signals have obvious fractal characteristics in the specified frequency bands, especially the electromagnetic interference signals. The fractal dimension box of specified frequency bands is close to the whole signal dimension box. The SVM network model with 23 fractal dimension vectors can be well used to recognize microseismic events, and the correct identification rate is 94%, with can meet the needs of the project site, but the efficiency of identification still needs further improvement.