基于梅尔倒谱系数的矿山复杂微震信号自动识别分类方法

为了实现矿山复杂微震信号的自动高效识别与分类，保证后续微震分析的时效性和准确性，运用梅尔倒谱系数法，将原始的4种微震信号(岩体破裂、爆破振动、电磁干扰和钻机凿岩)转化为梅尔标度上的非线性频谱，再转换到倒谱域上，结合其在时域上的差分得到1组24维的特征参数向量，利用这些特征参数向量训练构建各类事件对应的混合高斯隐马尔可夫识别模型，进而实现对微震信号的自动识别分类。研究结果表明:运用基于梅尔倒谱系数的微震信号识别分类方法对矿山实际微震数据进行测试，微震事件的识别分类准确率达到92.46%，具有较高的准确性，为实现微震监测系统的实时性分析提供了技术支持。

An automatic identification and classification method of complex microseismic signals in mines based on Mel-frequency cepstral coefficients

In order to realize the automatic and efficient identification and classification of complex microseismic signals in mines and ensure the timeliness and accuracy of subsequent microseismic analysis, the Mel frequency cepstral coefficients method was applied to convert four types of original microseismic signals (rock burst, blasting vibration, electromagnetic interference and rig drilling) into the nonlinear frequency spectrum on the MEL scale and then switch to the cepstrum domain, and a set of characteristic parameter vector with 24 dimensions was obtained combining with the difference in the time domain. The Gaussian mixture hidden Markov identification model corresponding to various events was constructed and trained by using these characteristics parameter vector, thus the automatic identification and classification of microseismic signals was realized. The experimental results showed that when testing the actual microseismic data of the mine by using the identification and classification method of microseismic signals based on the Mel frequency cepstral coefficients, the accuracy of identification and classification of microseismic events reached 92.46%, with a relatively high accuracy. It provides the technical support for realizing the real time analysis of microseismic monitoring system.