【作者】 沈跃；

【导师】 刘国海；

【作者基本信息】 江苏大学 ， 电力电子与电力传动， 2012， 博士

【摘要】 随着电网规模的扩大、电力信息化的发展和智能化电网的建设,电能质量问题日益成为电力部门和用户普遍关注的问题。电能质量数据的检测与识别是电能质量问题得以改善和控制的基础。电能质量数据的压缩能有效解决电能质量监测数据量大的问题,缓解系统数据存储和传输的负担。现有的信号检测与压缩方法均建立在香农采样定理基础上,这将导致海量采样数据,无论是当地存储还是传输给电力部门都将带来沉重的负担。近年来,一种新兴的压缩感知理论为数据采集技术带来了革命性的突破,得到了广泛关注。压缩感知采用非自适应线性投影来保持信号的原始结构,以远低于奈奎斯特频率进行采样,能通过数值最优化问题准确重构原始信号。本文针对以上研究背景,围绕压缩感知理论在电力系统谐波检测、电能质量扰动识别和电力系统数据压缩等方面的应用展开了一些有益的研究。为解决电力系统谐波检测问题在传统信号处理模式下检测过程的复杂性和冗余性,提出一种压缩采样正交匹配追踪的谐波信号检测算法。该方法首先基于压缩感知思想对原始谐波信号进行压缩采样,然后应用压缩采样正交匹配追踪算法直接对采样序列值进行谐波检测与分离。相比传统信号处理模式的采样—压缩—存储或传输—解压缩—信号检测过程,本文方法采样与压缩合二为一,并且无需将压缩信号完全重构后再进行信号检测,检测模式简化为压缩采样—存储或传输—信号检测。压缩采样检测方法只需要很少的信号采样点数即能对原始谐波信号精确检测,有利于减少模数采样设备的负担,节约中间变量的存储空间,能从压缩信号中直接检测出感兴趣的基波和各次谐波成份,并略去了解压缩重构过程,具有很好的应用前景。针对电能质量扰动识别方法中扰动特征提取方法依赖对象类别、识别算法复杂等问题,深入研究了基于压缩感知理论应用于电能质量扰动特征提取与扰动识别分类的可行性,提出了一种基于随机矩阵降维映射特征提取的电能质量扰动稀疏表示多分类方法。首先将扰动信号测试样本表示为训练样本集的过完备字典稀疏线性组合,然后使用随机测量矩阵获取测试样本降维特征量和稀疏表示感知矩阵,进而求取扰动信号测试样本的稀疏解,实现对电能质量扰动的稀疏表示多分类识别。该随机映射特征提取方法不依赖于电能质量扰动样本特性,适应于不同扰动信号类型,具有普适性；稀疏识别法与支持向量机相比具有更高的分类准确率,抗噪声鲁棒性更强,对于多分类情况无需由二分类器构造。仿真分析和实验结果表明该方法能高效快速地提取电能质量扰动信号特征,在噪声环境中获取高精度的电能质量扰动分类识别率。为减少电能质量扰动数据的采样存储空间和传输数据量,提高电能质量监测系统的实时性,在深入研究压缩感知理论的基础上,给出一种基于压缩感知理论的电能质量数据采样压缩和自适应匹配追踪数据重建方法。首先采用随机测量矩阵实现采样与压缩并行的电能质量数据压缩过程,随机测量矩阵构造简单、运算快速,无需中间变量存储空间,不依赖于电能扰动信号特征,具有普适性；然后基于自适应匹配追踪算法重建原始数据,相比于正交匹配追踪等贪婪算法,无需已知稀疏度,具备自适应和正则化过程,运行时间短,能够实现精确重建。该方法突破传统数据压缩方法先采样后压缩的框架,将采样与压缩合并进行,少量采样即能很好地恢复原始电能质量信号,不仅能够降低对硬件的要求,而且提高压缩效率更多还原

【Abstract】 With the enlargement of the power grids, development of informationization and construction of smart grid, power quality issues have increasingly captured considerable attention from both utility companies and their customers. The detection and identification research of power quality disturbances could provide a basis for executing effective control strategies to solve electric pollution and improve power quality. The research and application of power quality data compression is of great importance for reducing the data storage burden and improving the transfer efficiency. Due to the limitation of Shannon sampling theorem, the existing signal acquisition and detection methods will lead to mass of sampled data and bring heavy burden during storage and transmission.In recent years, compressed sensing (CS) has attracted considerable attention. Compressed sensing using non-adaptive linear projection to maintain the original structure of the signal, far below the Nyquist frequency of sampling can perfectly reconstruct the original signal by the numerical optimization problem. Several key issues, such as harmonics detection, power quality disturbance classification and data compression based on compressed sensing theory are studied and analyzed in the dissertation.To solve complexity and redundancy of detected harmonic by the traditional signal processing mode, the compressed sensing matching pursuit harmonics detection method is presented. At first, original harmonics was sampled and compressed simultaneously based on compressed sampling theorem. Then harmonics components could be solved by a compressed sensing matching pursuit harmonics detection algorithm, which directly processes the sampling values obtained from compressed sensing signal, without reconstructing the signal itself. Compared with the traditional mode, the new method merged sampling and compression to one step and skipped the uncompress process and can directly detect the interested base wave and harmonic components from the small amount of signal sampling points from the original harmonic compressed signal.In order to resolve the disadvantage of feature extraction relying on object category and complex classification algorithm, a new approach combining random dimensionality reduction projection (RDRP) with sparse representation classification (SRC) is proposed to classify power quality disturbances. Random matrix dimensionality reduction projection feature extraction method is extremely efficient to generate and independent of the training dataset. Compared with support vector machine, the SRC algorithm doesn’t need combination of two-class classifiers for multiclass classification. Simulation and experiment results show that the proposed RDRP-based SRC method has a high classification accuracy rate under noisy circumstance.A power quality data compressive sampling and adaptive matching pursuit reconstruction algorithm based on compressed sampling theorem is presented to reduce power quality data storage/transfer burden and improve the real time property of power quality real-time monitoring system. At first, original power quality data was sampled and compressed simultaneously based on compressed sampling theorem. The sampling feature points obtained from random matrix projection method is extremely efficient to generate and independent of object characteristics. Then the proposed adaptive matching pursuit reconstruction algorithm could control the accuracy of reconstruction by both the adaptive process which chooses the candidate set automatically and the regularization process which gets the atoms in the final support set although the sparsity of the original signal is unknown. The proposed method breaks through the traditional framework of data compression by merging compression into sampling and could reconstruct the original power quality data from the small amount of signal sampling points from the compressed signal. It can not only reduce hardware requirements, but also increase the efficiency of compression.更多还原