基于改进HHT的10kV的XLPE电缆接头典型局部放电的辨别  

Identification of Representative Partial Discharges in 10kV XLPE Cable Joint Based on Improved Hilbert-Huang Transform

作　　者：秦榛 王睿 王金鑫 彭浩城 QIN Zhen;WANG Rui;WANG Jinxin;PENG Haocheng(Chongqing Electric Power Company Economic and Technological Research Institute,Chongqing 401120,China;School of Mechanical Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]国网重庆市电力公司经济技术研究院,重庆401120 [2]江苏大学机械工程学院,江苏镇江212013

出　　处：《电力信息与通信技术》2022年第5期95-102,共8页Electric Power Information and Communication Technology

基　　金：重庆电力设计院有限责任公司采购项目“电缆隧道自适应巡检综合在线监控系统方案设计服务”(SJY20200501-03)。

摘　　要：针对局放信号中单一的去噪方法去噪效果差以及经验模态分解(empirical mode decomposition,EMD)存在的模态混叠和其他缺陷,文章提出一种基于改进的希尔伯特黄变换(Hilbert-Huang transform,HHT)的去噪方法;针对神经网络所需样本多和计算量大的特点,提出对有高信息维度的边际谱进行方向梯度直方图(histogram of gradient,HOG)和灰度共生矩阵(gray level co-occurrence matrix,GLCM)特征提取后通过支持向量机(SVM)辨别的方法。局放信号中主要有窄带周期干扰和白噪声2种噪声干扰难以去除,窄带噪声能量集中在频域上,可以通过快速傅里叶变换(FFT)将其先去除,然后采用添加互补自适应白噪声的完整集合经验模态分解(complete ensemble empirical mode decomposition with complementary adaptive white noise,CEEMDCAN)和自适应阈值法结合的方法,不仅有效抑制了白噪声,而且消除了模态混叠,重构误差和计算量都更小。仿真结果表明,该去噪方法去噪效果明显,在少样本情况下,通过SVM快速辨别可以获得92.5%的高识别率。An improved denoising method of partial discharge(PD)signals based on Hilbert-Huang Transform is presented in this paper in view of the poor denoising effect of the previous single denoising method and the modal aliasing problems and other defects of the empirical mode decomposition(EMD).And considering that neural network requires many samples and has large amount of computation,this paper uses support vector machine(SVM)by extracting HOG and GLCM features from Hilbert marginal spectrum which has high information dimensions.There are mainly two kinds of noise interference in the PD signals which are difficult to be removed:periodic narrowband noise and white noise.Periodic narrowband noise can be eliminated by Fast Fourier Transform firstly because its energy is concentrated in the frequency domain.Then,the method combining complete ensemble empirical mode decomposition with complementary adaptive white noise(CEEMDCAN)and adaptive threshold processing not only effectively suppresses the white noise but also eliminates the mode aliasing problems and reduces the calculation and reconstruction error.The simulation results show that the denoising effect of the proposed method is obvious.In the case of small samples,the high recognition rate of 92.5%can be obtained by fast identification with SVM.

关 键 词：局部放电 希尔伯特黄变换(HHT) 去噪 边际谱 支持向量机 

分 类 号：TM855[电气工程—高电压与绝缘技术]