基于ICEEMDAN和MC-CNN的矿山声发射信号识别分类方法*

为精准识别地下矿山声发射事件，采用基于改进的完全集合经验模态分解模型(ＩＣＥＥＭＤＡＮ)和多通道卷积神经网络（ＭＣ－ＣＮＮ）模型对声发射信号进行处理后得到分量图，根据各通道输入分量峭度值赋予不同权重，并利用卷积神经网络对输入数据进行训练，最终采用五折交叉实验方法验证该分类识别方法的可行性及有效性。结果表明:基于ICEEMDAN和MC-CNN模型分类识别正确率为97.64%，与其他传统识别方法相比能精准有效地对地下矿山声发射信号进行识别分类，显著提高卷积神经网络的波形识别正确率。研究结果可为地下矿山声发射事件识别分类提供借鉴。

Recognition and classification methods of mine acoustic emission signals based on ICEEMDAN and MC-CNN

In order to accurately identify the acoustic emission events in underground mines,the acoustic emission signals are processed based on the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) and multi-channel convolutional neural network (MC-CNN) model,and then intrinsic mode function are obtained.Different weights are given according to the kurtosis values of the input components in each channel,and the input data are trained by the convolutional neural network.Finally,the feasibility and effectiveness of the classification and identification method are verified by the Five-fold cross experiment method.The results show that the classification recognition accuracy based on ICEEMDAN and MC-CNN model is 97.64%.Compared with other traditional recognition methods,it can accurately and effectively classify the acoustic emission signals of underground mines,and significantly improve the waveform recognition.