基于小波散射分解变换的煤矿微震信号智能识别  

Intelligent recognition of coal mine microseismic signal based on wavelet scattering decomposition transform

作　　者：樊鑫 程建远 王云宏 栗升 段建华 王盼[1,2] FAN Xin;CHENG Jianyuan;WANG Yunhong;LI Sheng;DUAN Jianhua;WANG Pan(College of Geology and Environment,Xi’an University of Science and Technology,Xi’an 710054,China;Xi’an Research Institute,China Coal Technology and Engineering Group,Xi’an 710077,China)

机构地区：[1]西安科技大学地质与环境学院,陕西西安710054 [2]中国煤炭科工集团西安研究院有限公司,陕西西安710077

出　　处：《煤炭学报》2022年第7期2722-2731,共10页Journal of China Coal Society

基　　金：国家自然科学基金资助项目(42074175);国家重点研发计划资助项目(2018YFC0807804)。

摘　　要：微震信号是一种典型的时变非平稳信号,微震监测系统在高噪声环境下采集信号的信噪比偏低,影响了微震事件的识别准确率和精度,现有的微震事件识别方法仍然存在低速率、高时延、低精度等问题。以小波理论为基础,提出利用小波散射分解变换提取微震事件和噪声信号的特征,计算2类信号的特征系数,并构成相应的特征矩阵;基于微震信号低频特性、不可预知性和突发瞬态性的特点,对比分析了小波散射分解变换结构的主要参数:时不变尺度、散射分解次数和质量因子对特征矩阵维数的影响;通过调整参数的大小有效控制特征矩阵维数,提高运算效率,最终选择出最优的特征矩阵,利用SVM分类对360组煤矿微震信号进行分类识别。实验结果表明:①小波散射系数构成的特征矩阵可以有效区分微震事件信号与噪声信号,且事件信号特征系数变化趋势具有突变性,噪声信号特征系数变化趋势表现为无序性;②采用3次小波散射分解变换构成的特征矩阵,可有效表征微震事件信号与噪声信号的差异特征,提高程序运算效率,而随着变换次数增加,事件信号的低频部分更明显;③在时不变尺度和变换次数确定的情况下,为得到最优特征矩阵,各阶小波散射变换的质量因子选择不宜过大。Microseismic signal is a typical time-varying nonstationary signal,while the collected data are often disturbed by strong noise,seriously affecting the recognition efficiency of microseismic events and the accuracy of source location.Existing traditional recognition methods still have problems with a low rate,high delay and low accuracy.Therefore,based on wavelet theory,the wavelet scattering decomposition transformation method is proposed to analyze and extract the features of microseismic events and noise events signals,calculate the feature coefficients of the two signals,and form the corresponding feature matrix.Based on the characteristics of low-bandwidth,unpredictability and burst transient of microseismic signals,the effects of time-invariant scale,the number of scattering decomposition and quality factors on the dimension of the feature matrix are compared and analyzed by processing the real data.By adjusting the size of the parameters,the dimension of the feature matrix is effectively controlled to improve the operation efficiency.Finally,the optimal feature matrix is selected for the training of the machine learning classification model.The experimental results show that the feature matrix composed of the wavelet scattering coefficients can effectively distinguish the event signals from the noise signals,and the feature coefficients change trend of the event signals is mutant,and the noise signals feature coefficients change trend appears as a disorder.The feature matrix composed of three times wavelet scattering decomposition transformation can effectively represent the different characteristics of microseismic event signals and noise signals and improve programming efficiency,and the low-frequency part of event signal is more obvious with the number of transformations increasing.In the case where the time-invariant scale and the number of transformations are determined,in order to obtain the optimal feature matrix,the quality factor selection of the wavelet scattering transformation of each order sho

关 键 词：微震监测 小波散射分解 信号分析 特征提取 智能识别 

分 类 号：TD326[矿业工程—矿井建设] TN911.7[电子电信—通信与信息系统]