电力变压器局部放电信号的模式识别及传播特性

发布时间：2018-06-30 06:55

本文选题：局部放电 + 阈值去噪　；参考：《中国矿业大学》2017年硕士论文

【摘要】：局部放电既是电力变压器绝缘故障的重要诱发因素,又是绝缘潜伏性故障的高灵敏度表现形式。因此,局放信号的检测、局放故障模式识别及故障定位的研究,一直备受重视。本文制作了代表气隙放电、空气中电晕放电、油中沿面放电和空气中沿面放电四种典型局放故障的实验模型,搭建了实验平台,采集了典型局放故障的原始信号,进行了局放故障的模式识别研究。首先,针对局部放电信号非平稳、非线性强、易受干扰等特性,结合经验模态分解法,提出了基于完全经验模态分解和总体经验模态分解(CEEMD-EEMD)的局部放电阈值去噪新方法。利用新方法和常规的小波阈值去噪算法对仿真与实测局部放电信号进行去噪,去噪结果表明,基于CEEMD-EEMD的局部放电阈值去噪方法去噪效果更优,验证了该方法的有效性,为局部放电故障的特征提取与模式识别奠定了基础。其次,根据局部放电相位分布(PRPD)模式绘制了不同绝缘故障类型的二维图谱、三维图谱以及灰度图,利用主成分因子分析法将36维统计特征量降到8维具有物理意义的特征量。为了最大限度提取到局放故障最本质的信息,尝试采用自回归模型算法提取局部放电信号特征,通过BP神经网络和SVM对这两类特征进行分类识别,结果表明基于自回归模型系数特征的局部放电识别率均较高,说明局放故障的自回归系数特征要优于统计特征量。再次,利用超球面支持向量机对不同绝缘故障局部放电类型进行模式识别。局部放电信号检测复杂,对应故障类型多样,局部放电样本数目有限且特征量呈非线性,使得BP神经网络和SVM的识别率较低。本文基于自回归系数特征,采用经过粒子群优化的超球面支持向量机对不同绝缘故障类型的局部放电进行模式识别,识别率高,这对提高局部放电模式识别率具有一定的指导意义。最后,应用时域有限差分方法的全波三维电磁场仿真软件(XFDTD),研究局部放电产生的电磁波在变压器绕组中的传播特性。利用XFDTD构建了变压器绕组仿真模型,利用脉冲信号源来模拟局部放电信号,设定假想监测点来测量电磁波衰减情况,仿真结果表明:当电磁波通过绕组时,其幅值衰减率在65%左右;当绕组线饼宽度变大时,通过绕组电磁波的幅值衰减程度越严重,电磁波传播的时延差也会变大。
[Abstract]:Partial discharge (PD) is not only an important inductive factor of insulation fault of power transformer, but also a high sensitive form of insulation latent fault. Therefore, the research of PD signal detection, PD fault pattern recognition and fault location has been paid more attention. In this paper, the experimental models of four typical partial discharge faults, representing air gap discharge, corona discharge in air, surface discharge in oil and surface discharge in air, are made, and the experimental platform is built, and the original signals of typical partial discharge faults are collected. The pattern recognition of PD fault is studied. Firstly, a new method of partial discharge threshold de-noising based on complete empirical mode decomposition and total empirical mode decomposition (CEEMD-EEMD) is proposed, which is based on the characteristics of non-stationary, strong nonlinear and easily disturbed partial discharge signals. The new method and the conventional wavelet threshold denoising algorithm are used to Denoise the simulated and measured PD signals. The results show that the denoising effect of the PD threshold denoising method based on CEEMD-EEMD is better, and the effectiveness of the proposed method is verified. It lays a foundation for feature extraction and pattern recognition of partial discharge faults. Secondly, according to the partial discharge phase distribution (PRPD) pattern, the two-dimensional, three-dimensional and gray-scale maps of different insulation fault types are drawn, and the 36-dimensional statistical characteristic is reduced to 8-dimensional physical characteristic by principal component factor analysis (PCA). In order to extract the most essential information of PD fault to the maximum extent, the autoregressive model algorithm is used to extract PD signal features, and these two kinds of features are classified and recognized by BP neural network and SVM. The results show that the recognition rate of partial discharge based on the coefficients of autoregressive model is higher than that of statistical feature. Thirdly, the hyperspherical support vector machine is used to identify the partial discharge types of different insulation faults. The detection of partial discharge signals is complicated, the corresponding fault types are various, the number of partial discharge samples is limited and the characteristic quantity is nonlinear, which makes the recognition rate of BP neural network and SVM low. Based on the feature of autoregressive coefficient, a hyperspherical support vector machine based on particle swarm optimization (PSO) is used to identify the partial discharge (PD) of different insulation fault types, and the recognition rate is high. It has certain guiding significance to improve partial discharge pattern recognition rate. Finally, the full wave 3D electromagnetic field simulation software (XFDTD) is used to study the propagation characteristics of electromagnetic waves generated by partial discharge in transformer windings. The simulation model of transformer winding is constructed by using XFDTD, the partial discharge signal is simulated by pulse signal source, and the attenuation of electromagnetic wave is measured by setting the imaginary monitoring point. The simulation results show that when the electromagnetic wave passes through the winding, The attenuation rate of the amplitude is about 65%, and when the width of the winding wire-cake becomes larger, the amplitude attenuation degree of the electromagnetic wave passing through the winding becomes more serious, and the delay difference of the electromagnetic wave propagation will also become larger.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM41

【参考文献】