电能质量复合扰动识别方法研究

发布时间：2018-11-07 09:57

【摘要】：近几年,智能电网的研究成为电力系统的一个热点问题,而保障优质的电能质量是智能电网研究的重点问题。另一方面,大量电力电子设施的广泛使用,新能源等技术的应用,都需要高质量的电能提供保障。所以,识别出电能质量信号中的扰动信息不仅有利于检测出优劣的电能质量,而且还能减少或者控制由电能质量扰动产生的各种问题。在实际生活中出现的扰动并不只是单一的扰动,而是经常出现几种扰动共存的情况。因此,识别出扰动是保障优质的电能质量的基础。本课题重点是围绕复合扰动的特征提取和分类识别两部分展开探究。在特征提取方面,本文主要是应用S变换和小波变换提取特征量。本文在探究复合扰动的提取特征时,用S变换对扰动作深入探究,提出一种提高时间和频率分辨率的S算法,提取每种扰动的改进的S矩阵的每列最大幅值的最大值、每列最大幅值的最小值和工频幅值的均值三个特征量作为一部分特征量;对扰动信号进行小波变换,提取扰动信号每层能量的差值作为另一部分,加上改进S变换提取的一部分特征量作为总的特征量。在分类方面,应用支持向量机识别出不同的扰动。其中,高斯核函数是其辨识出扰动信号的关键因子。本文对高斯核函数进行改进,引入幅度调节参数和径向宽度调节参数,提高了电能质量复合扰动的识别准确率;对于分类器中的参数选择难的问题,用粒子群进行参数寻优,并且深入研究粒子群,提出了指数型的惯性权重,快速准确的求取参数的最优组合,提高了扰动识别的准确率。仿真结果显示,利用小波算法和提高时频分辨率的S算法获取特性向量用到支持向量机中,得到的识别准确率比小波变换和S变换提取的特征量进行识别的准确率提高了3.7839%,比小波变换提取的特征量的识别准确率提高了7.5758%;利用基于幅度调节和径向宽度调节的高斯核函数算法,提高了支持向量机分类器的识别准确率,降低了计算复杂度,使支持向量的个数变少,其整体识别准确率比支持向量机的提高了1.8182%;利用指数型惯性权重的粒子群算法求取改进的支持向量机中的参数的最优值,得到的识别准确率比粒子群得到的结果提升了0.3788%。
[Abstract]:In recent years, the research of smart grid has become a hot issue in power system, and the guarantee of high quality power quality is the key issue in the research of smart grid. On the other hand, the widespread use of a large number of power electronic facilities and the application of new energy technologies require high quality electrical energy to provide protection. Therefore, identifying the disturbance information in the power quality signal is not only helpful to detect the power quality, but also can reduce or control all kinds of problems caused by the power quality disturbance. The disturbance in real life is not only a single disturbance, but also the coexistence of several disturbances. Therefore, the identification of disturbances is the basis for ensuring high quality power quality. This thesis focuses on feature extraction and classification recognition of complex disturbances. In feature extraction, this paper mainly uses S transform and wavelet transform to extract feature quantity. In this paper, an S algorithm is proposed to improve the resolution of time and frequency in order to extract the maximum of the maximum value in each column of the improved S-matrix of each disturbance. The minimum value of the maximum value of each column and the mean value of the power frequency amplitude are taken as part of the eigenvalues. Wavelet transform is applied to the disturbance signal, the difference of energy in each layer is extracted as the other part, and a part of the characteristic quantity extracted by the improved S transform is taken as the total characteristic quantity. In classification, support vector machines (SVM) are used to identify different disturbances. Among them, Gao Si kernel function is the key factor to identify disturbance signal. In this paper, Gao Si kernel function is improved, amplitude adjustment parameter and radial width adjustment parameter are introduced to improve the accuracy of power quality complex disturbance identification. For the problem of difficult parameter selection in classifier, the particle swarm optimization is used to optimize the parameters, and the particle swarm optimization is deeply studied. The inertial weight of exponential type is put forward, the optimal combination of parameters is obtained quickly and accurately, and the accuracy of disturbance identification is improved. The simulation results show that the wavelet algorithm and the S algorithm to improve the time-frequency resolution are used to obtain the characteristic vector in the support vector machine. The recognition accuracy is 3.7839 higher than that of wavelet transform and S-transform, and 7.5758% higher than that of wavelet transform. Using Gao Si kernel function algorithm based on amplitude adjustment and radial width adjustment, the recognition accuracy of support vector machine classifier is improved, the computational complexity is reduced, and the number of support vectors is reduced. The overall recognition accuracy is 1.8182 higher than that of support vector machine. The particle swarm optimization algorithm of exponential inertia weight is used to obtain the optimal value of the parameters in the improved support vector machine, and the recognition accuracy is improved by 0.3788 compared with the result obtained by the particle swarm optimization.
【学位授予单位】：东北电力大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM76;TP18

【参考文献】