融合CEEMDAN-GMM和小波阈值方法的爆炸冲击波降噪算法  

Explosion Shock Wave Denoising Algorithm Based on Fusion of CEEMDAN-GMM and Wavelet Threshold Method

作　　者：杨志飞 崔春生[1,2] 杜桂云 刘双峰 赵海霞[3] YANG Zhifei;CUI Chunsheng;DU Cuiyun;LIU Shuangfeng;ZHAO Haixia(State key Laboratory of Dynamic Measurement Technology,North University of China,Taiyuan 030051,Shanxi,China;School of Electrical and Control Engineering,North University of China,Taiyuan 030051,Shanxi,China;School of Environment and Safety Engineering,North University of China,Taiyuan 030051,shanxi,China)

机构地区：[1]中北大学省部共建动态测试技术国家重点实验室,山西太原030051 [2]中北大学电气与控制工程学院,山西太原030051 [3]中北大学环境与安全工程学院,山西太原030051

出　　处：《弹箭与制导学报》2023年第5期80-86,共7页Journal of Projectiles,Rockets,Missiles and Guidance

基　　金：中北大学重点实验室开放研究基金(DXMBJJ2021-01);中国应用物理化学科学与技术实验室项目(614260209)资助。

摘　　要：针对实测爆炸冲击波信号中存在大量噪声信号以及自适应白噪声完整集合经验模态分解(CEEMDAN)方法在降噪过程中容易出现信号失真的问题,提出了一种基于CEEMDAN方法和高斯混合模型(GMM)的爆炸冲击波降噪算法。该算法首先将原始爆炸冲击波数据通过CEEMDAN方法分解得到若干个本征模态分量IMF,并利用GMM聚类算法将IMF分量分为多个类别。通过计算不同IMF分量的方差贡献率校核聚类结果,剔除高频噪声类别的IMF分量。对受噪声干扰的IMF分量,采用小波阈值方法进行滤波去除噪声。最后,将滤波后的IMF分量与未受噪声干扰的IMF分量一起重构,并获得降噪后的爆炸冲击波信号。实验结果表明:CEEMDAN-GMM与小波阈值算法信噪比值RSNR和均方根误差值MRMSE分别比CEEMDAN方法和CEEMDAN-小波阈值算法提高了10.19 dB和2.99 dB,降低了0.42×10^(-4)和0.14×10^(-4)。In response to the large number of noise signals in the measured explosion shock wave signal and the problem of sig-nal distortion that may occur during the noise reduction process using the CEEMDAN method.A noise reduction algorithm for explosive shock waves based on adaptive white noise complete set empirical mode decomposition(CEEMDAN)and Gaussian mixture model(GMM)is proposed.This algorithm first decomposes the original explosion shock wave data into several intrin-sic mode components(IMF)through CEEMDAN,and uses the GMM clustering algorithm to classify the IMF components into multiple categories.Verify the clustering results by calculating the variance contribution rates of different IMF components,and eliminate IMF components in high-frequency noise categories.For IMF components affected by noise interference,wavelet thresholding method is used to filter and remove noise.Finally,the filtered IMF component is reconstructed together with the IMF component that is not affected by noise to obtain the denoised explosion shock wave signal.The experimental results show that the SNR and RMSE values of CEEMDAN-GMM and wavelet thresholding algorithms are 10.19 dB and 2.99 dB higher than CEEMDAN and CEEMDAN-Wavelet thresholding algorithms,respectively,with a decrease of 0.42×10^(-4) and 0.14×10^(-4).

关 键 词：爆炸冲击波信号 CEEMDAN 降噪 GMM 

分 类 号：O384[理学—流体力学]