基于小波变换的电能质量检测与仿真分析

发布时间：2018-10-23 17:10

【摘要】：近年来，电能质量问题已引起电力部门以及用户的广泛关注。电能质量检测是监督、改善电能质量的一个非常必要的前提，对保证电力系统的安全经济运行以及用电安全具有重要的理论和实际意义。本文重点研究了常见电能质量扰动信号的时间定位和分类问题。本文首先对国内外电能质量检测方面的研究进行总结，从不同角度描述了电能质量的定义以及分类方法，分析总结了电能质量的相关国家标准以及电能质量检测新要求和发展趋势，并给出了7种常见电能质量扰动的数学模型。然后详细地介绍了小波理论及其性质，探讨了小波在电能质量检测中的应用。重点研究了基于小波变换的电能质量扰动信号奇异性检测原理和分类特征向量的提取方法。通过仿真分析，在三维视角上直观地呈现出所提取特征向量的区分空间，验证了提取的分类特征向量的有效性。电能质量扰动信号的检测与定位为分析扰动产生的原因提供依据。文中提出了一种基于复小波的电能质量扰动检测与定位方法。该方法利用离散复小波变换，提取扰动信号的复小波系数的幅值和相位信息，再利用幅值和相位的复合信息实现对5种暂态电能质量扰动信号的时间定位。在噪声条件下该方法仍然适用，但是，当短时电能质量扰动的起止点发生在信号的幅值过零点附近时该方法将失效。针对这种情况，提出了一种辅助定位方法——对信号作小波分解与重构，获取信号低频波形，再对其使用复小波变换。仿真表明，该方法能在噪声条件下实现对电能质量扰动信号的快速准确定位。准确的识别和分类电能质量扰动对分析和综合治理电能质量问题具有重要意义。文中提出了一种基于小波和改进神经树的电能质量扰动分类方法，该方法利用小波分解扰动信号到各个频带，在基频频带、谐波频带和高频带上分别计算其能量值和小波系数熵作为特征值，另计算基波频带扰动过程的均方根作为特征的补充，融合能量、熵和均方根值作为扰动分类的特征向量，规范化后输入到改进神经树分类器进行训练和分类，改进神经树分类器是由神经网络和决策树及其分类规则构成。仿真表明，该方法提取特征值的计算量小且融合后的特征向量能够很好体现不同扰动信号之间的差异信息，，构造的改进神经树分类器结合了神经网络和决策树在模式分类中各自的优点，结构简单且表现出良好的收敛性、全局最优性和泛化性，且分类准确率较高，能够有效地识别7种常见的电能质量扰动。
[Abstract]:In recent years, the power quality problem has aroused the widespread concern of the electric power department as well as the user. Power quality detection is a very necessary prerequisite for monitoring and improving power quality. It is of great theoretical and practical significance to ensure the safe and economical operation of power system and the safety of power consumption. This paper focuses on the time localization and classification of common power quality disturbance signals. Firstly, this paper summarizes the research on power quality detection at home and abroad, and describes the definition and classification of power quality from different angles. This paper analyzes and summarizes the relevant national standards of power quality, the new requirements and development trend of power quality detection, and gives seven mathematical models of power quality disturbances. Then the wavelet theory and its properties are introduced in detail, and the application of wavelet in power quality detection is discussed. The singularity detection principle of power quality disturbance signal based on wavelet transform and the extraction method of classification feature vector are studied. Through simulation analysis, the distinguishing space of the extracted feature vectors is presented intuitively from the three-dimensional perspective, and the validity of the extracted feature vectors is verified. The detection and location of power quality disturbance signal provide basis for analyzing the cause of disturbance. In this paper, a power quality disturbance detection and localization method based on complex wavelet is proposed. In this method, the amplitude and phase information of complex wavelet coefficients of disturbance signals are extracted by discrete complex wavelet transform, and the time localization of five kinds of transient power quality disturbance signals is realized by using the composite information of amplitude and phase. The method is still applicable under the noise condition, but it will fail when the starting and ending point of the short term power quality disturbance occurs near the zero crossing point of the signal amplitude. In order to solve this problem, an auxiliary localization method is proposed, which is to decompose and reconstruct the signal by wavelet transform, obtain the low frequency waveform of the signal, and then use complex wavelet transform. Simulation results show that the proposed method can locate the power quality disturbance signals quickly and accurately under noise conditions. Accurate identification and classification of power quality disturbances is of great significance in analyzing and synthesizing power quality problems. In this paper, a power quality disturbance classification method based on wavelet and improved neural tree is proposed. The energy value and wavelet coefficient entropy of harmonic band and high frequency band are calculated as eigenvalues respectively, and the root mean square (RMS) of fundamental frequency band perturbation process is calculated as the supplement of the feature, and the energy, entropy and RMS value are used as eigenvectors of disturbance classification. The improved neural tree classifier is composed of neural network, decision tree and its classification rules. Simulation results show that the proposed method can well represent the difference information between different disturbance signals, and the computation of the extracted eigenvalues is small and the fused Eigenvectors can well reflect the difference between different disturbance signals. The improved neural tree classifier combines the advantages of neural network and decision tree in pattern classification. It has the advantages of simple structure, good convergence, global optimality and generalization, and high classification accuracy. It can effectively identify seven common power quality disturbances.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM711

【参考文献】