油纸绝缘典型缺陷局部放电特征提取与模式识别研究

发布时间：2018-05-24 08:11

本文选题：局部放电 + 特征提取　；参考：《中国矿业大学》2015年硕士论文

【摘要】：电力变压器是电力系统中最关键的设备之一,其安全可靠运行对电网至关重要。而局部放电是引起电力变压器绝缘老化和绝缘故障的主要因素,不同局部放电类型对绝缘损害的程度不同,其形成的机理也各有差异。因此,针对不同的局部放电类型进行模式识别,对快速诊断变压器绝缘状态及辨识故障位置具有重要意义。本文深入研究分析了局部放电机理及危害,通过搭建局部放电实验平台对4种典型的缺陷模型进行试验,研究各种类型的放电图谱并对提取统计特征参数进行定量分析。本文主要内容如下:首先,根据CMII标准制作了4种典型的放电模型,用于模拟不同类型的油纸绝缘变压器局部放电故障缺陷。通过在高压实验室搭建局部放电实验平台及检测电路,利用脉冲电流法检测放电产生的脉冲信号,并分析研究了不同种类缺陷模型的放电特性,总结其放电机理及波形特征。其次,在研究小波变换消噪方法的基础上,采用db8小波对实测局部放电信号进行5层分解,滤除含有的噪声干扰。根据局部放电相位分布PRPD模式,绘制得到各放电类型的二维、三维谱图,并提取能够表征图谱特性的30个统计特征,为后续量化分析识别各放电类型提供依据。第三,应用主成分分析和核主成分析法对提取的30个统计特征参数进行降维处理,并将两者的降维效果进行对比。结果表明:运用主成分分析法降维后得到9个新特征量,而利用核主成分分析法降维后只需6维数据特征,且前三个主成分的累计贡献率已达到降维目的。降维后的特征量明显减少,综合变量保留了原数据的特征信息,为后续多分类支持向量机对局部放电类型识别奠定了基础。第四,构造多分类优化参数支持向量机SVM分类器用于不同类型局部放电的识别。首先采用网格搜索算法实现支持向量机的参数优化,将降维后的局部放电综合特征作为分类特征量,并将训练集采用5折交叉验证法寻找最优训练SVM模型;然后,结合M-ary分类思想,将支持向量机的两类分类问题扩展为多类分类领域;最后,将测试数据分别输入到未优化SVM和优化参数多分类SVM模型中进行分类测试识别。比较各分类器的识别准确率及可靠性。实验结果表明,该方法计算速度快,且获得了较好地识别效果,适用于局部放电类型识别。
[Abstract]:Power transformer is one of the most important equipments in power system. Partial discharge (PD) is the main factor that causes insulation aging and insulation failure of power transformer. Different types of partial discharge have different degree of insulation damage, and their formation mechanism is different. Therefore, pattern recognition for different partial discharge types is of great significance for rapid diagnosis of transformer insulation and identification of fault location. In this paper, the mechanism and harm of partial discharge (PD) are deeply studied and analyzed. Four typical defect models are tested by setting up a PD experimental platform, and the discharge patterns of various types are studied and the statistical characteristic parameters extracted are quantitatively analyzed. The main contents of this paper are as follows: firstly, four typical discharge models are made according to CMII standard to simulate different types of partial discharge faults of oil-paper insulated transformers. By setting up a partial discharge experimental platform and detecting circuit in the high voltage laboratory, the pulse current method is used to detect the pulse signal generated from the discharge, and the discharge characteristics of different kinds of defect models are analyzed, and the discharge mechanism and waveform characteristics are summarized. Secondly, based on the research of wavelet transform denoising method, db8 wavelet is used to decompose the measured PD signal in five layers to filter the noise interference. According to the PRPD mode of partial discharge phase distribution, the 2D and 3D spectra of each discharge type are drawn, and 30 statistical features which can characterize the characteristics of the spectrum are extracted, which provide the basis for the subsequent quantitative analysis and identification of each discharge type. Thirdly, principal component analysis (PCA) and kernel principal component analysis (KPCA) are used to reduce the dimensionality of 30 statistical characteristic parameters, and the effects of them are compared. The results show that nine new characteristic quantities are obtained by using principal component analysis (PCA), but only 6 dimensional data features are needed after dimensionality reduction by kernel principal component analysis (KPCA), and the cumulative contribution rate of the first three principal components has reached the goal of dimensionality reduction. After dimensionality reduction, the feature quantity is obviously reduced, and the feature information of the original data is preserved by the comprehensive variables, which lays the foundation for the recognition of partial discharge types by the subsequent multi-classification support vector machines. Fourthly, SVM classifier based on multi-classification optimization parameter support vector machine (SVM) is constructed to identify different types of partial discharges (PD). Firstly, the parameter optimization of support vector machine is realized by using mesh search algorithm. The feature of partial discharge synthesis after dimensionality reduction is taken as the classification feature quantity, and the training set is used to find the optimal training SVM model by 5 fold cross validation method. Combined with the idea of M-ary classification, the two kinds of classification problems of support vector machine are extended to multi-class classification field. Finally, the test data are input into the unoptimized SVM model and the multi-classification SVM model with optimized parameters for classification and identification. The recognition accuracy and reliability of each classifier are compared. The experimental results show that the proposed method has high speed and good recognition effect and is suitable for partial discharge type recognition.
【学位授予单位】：中国矿业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM855

【参考文献】