基于改进CEEMDAN-DCNN的声发射源识别分类方法

声发射源的准确分类识别是声发射地压监测预报预警研究的重要基础。针对矿山井下围岩体声发射事件信号和采掘作业噪声信号分类识别问题，提出了一种基于改进完备总体经验模态分解和深度卷积神经网络（DCNN）的智能识别分类方法。首先，对信号进行改进CEEMDAN降噪处理，即利用相关性系数阈值和排列熵（PE）阈值剔除伪分量和噪声分量；然后，利用DCNN对降噪后的信号自动提取高维特征；最后，将特征用于softmax分类器分类识别，实现智能化井下信号源多分类。研究表明：改进CEEMDAN能够有效剔除伪分量及噪声分量；相比其他机器学习方法，改进CEEMDAN-DCNN方法具有准确率高和稳定性较好等优点。信号源识别分类方法研究为地压监测预警预报提供了重要的基础数据，准确的灾害预警预报可为矿山井下作业人员和设备提供安全保障。

Identification and Classification Method of Underground AE Source Based on Improved CEEMDAN-DCNN

Accurate classification and identification of acoustic emission sources is an important basis for the study of acoustic emission ground pressure monitoring， forecasting and early warning.Aiming at the clas-sification and identification of acoustic emission event signals and mining operation noise signals of surrounding rock masses in underground mines， an intelligent recognition and classification method based on improved complete ensemble empirical mode decomposition and deep convolutional neural network（DCNN）was proposed.Firstly，the signal was decomposed by CEEMDAN， the decomposed IMF components were screened， and the components greater than the permutation entropy threshold or less than the correlation coefficient threshold were removed， and the residual IMF components were reconstructed to obtain the denoised waveform.Then， the DCNN method was used to automatically extract high-dimensional features from the denoised waveform.Finally， the features were used for classification and recognition of softmax classifier to realize intelligent multi-classification of underground signal sources.The results of this research show that：（1）Aiming at the difficulty of multi-classification of waveforms received by acoustic emission monitoring equipment，a waveform classification and recognition method based on improved CEEMDAN-DCNN is proposed.Combined with the advantages of improved CEEMDAN’s advantages of adaptive analysis，pro-cessing of nonlinear and non-stationary signals and the ability of DCNN to automatically extract high-dimensional features， the intelligent multi-classification of underground signal sources is realized.（2）In order to verify the advantages of the improved CEEMDAN algorithm， the simulation signal is constructed to simulate the acoustic emission signal of surrounding rock mass containing noise signal， and the background noise component and pseudo component are eliminated by a joint threshold.The results show that the improved CEEMDAN algorithm can eliminate noise signals and some false components， and retain the essential characteristics of the signal.（3）Through the test， the accuracy of waveform classification based on the improved CEEMDAN-DCNN method in this paper reaches 97.12%. Compared with the traditional SVM， ANN， and CNN methods， the accuracy of waveform classification is higher and the stability is better. The accuracy of DCNN classification and recognition is improved dueing to the signal preprocessed by improved CEEMDAN.（4）The waveform recognition and classification method in this paper can accurately identify the acoustic emission events of surrounding rock masses and non-surrounding rock masses， provide reliable basic research data for ground pressure monitoring and early warning models， and increase the accuracy of ground pressure monitoring and safety early warning and forecasting.