矿山微震波形特征自动模式识别算法研究

Research on automatic pattern recognition algorithm of micro-seismic waveform characteristics in mines

微震监测技术在矿山的应用需求日益增加,但还没有实现对有效信号与噪音信号的自动识别,严重制约了其应用效果与推广普及。深入研究与梳理了矿山强噪音环境下各主要待识别模式类,分别为凿岩、无轨设备行驶、溜井倒矿、电磁干扰、爆破与有效信号共六类,将有效信号模式类分为小能量事件与大能量事件2个子集,详细研究了各模式类的发生机制。通过收集各模式类的大量样本,研究抽取得到上述6类模式类的识别特征分别为:波形间隔时间Δt典型识别特征、波形持续时间t_(c)典型识别特征、总持续时间长度与内有单独事件个数联合的综合典型识别特征、t_(c)或主频f_(主)的典型识别特征、t_(c)与主频f_(主)联合的一般综合识别特征、t_(c)典型识别特征与排除法。采用预处理方法计算与统计得到了各识别特征值的分布概率。构建了识别性能优异的判决函数,建立了矿山微震波形特征自动模式识别算法。依托上述算法开发出了模式自动识别软件,通过在典型矿山的现场测试,有效信号的识别准确率为90.8%,现场应用效果好,实现了预期目标。

The application of the micro-seismic monitoring technology in mines is increasing,but the automatic identification of effective signals and noise signals has not been realized,which seriously restricts its application and popularization. Six main pattern categories in the mine strong noise environment,including drilling,trackless equipment running,mine shaft dumping,electromagnetic interference,blasting and effective signal,are investigated,and the effective signal patterns are divided into two subsets of small-energy events and large-energy events. The generation mechanism of each pattern class is studied in detail. By collecting a large number of samples of each pattern class,the identifying characteristics extracted for the above six pattern classes are respectively the waveform interval time,the waveform duration,the combination of the total duration with the number of individual events,the waveform duration or the main frequency,the combination of the waveform duration and the main frequency,and the waveform duration and the exclusion method. The distribution probabilities of the recognition characteristic values are calculated and counted by using the pre-processing method. A decision function with excellent recognition performance is constructed,and an automatic pattern recognition algorithm of mine micro-seismic waveform characteristic is established. Based on the above algorithms,a software of automatic pattern recognition is developed. Through the field test in a typical mine,the recognition accuracy of the effective signal of the developed algorithm is 90.8%,showing a good field application.