节点文献

电能质量扰动信号检测与识别算法研究

Detection and Recognition Algorithms Study on Power Quality Disturbances

【作者】 赵静

【导师】 钱清泉; 何正友;

【作者基本信息】 西南交通大学 , 电力系统及其自动化, 2011, 博士

【摘要】 目前,我国正处在工业化、城镇化加快推进的重要时期,这一时期电力需求一直保持较快的增长态势,给现代大电网互联的迅速发展以及电力系统规模的日益扩大带来了发展契机。但随之而来的各种电力电子设备、冲击性与非线性负荷的大量投入,导致供电质量出现电压暂降、电压暂升、电压波动、谐波、电压缺口、脉冲与振荡暂态干扰等一系列电能质量问题。因此,对电能质量信号进行实时监控与智能化分析,搭建电能质量扰动分析的理论框架,不仅可为电能质量扰动的检测、识别与扰动源定位等问题提供一定的解决思路和辅助方法,还为电能质量的管理与治理提供辅助决策,具有重要的现实意义。基于此,本文在分析稳态扰动、暂态扰动以及混合扰动的基础上,进行了电能质量扰动的检测、识别以及定位研究。在扰动检测方面,论文首先提出了一种新的基于数学形态学与熵理论的电能质量扰动信号检测方案。该方案利用广义形态滤波器对信号进行滤波处理,使其在滤除噪声的同时尽量保持信号原有扰动特征;再利用自定义的差分熵对信号突变程度进行度量,一般来说,信号突变处的熵值会比较大,由此可定位扰动开始与结束的时刻。其次,论文定义了一种新的形态非抽样小波(Morphological Undecimated Wavelet, MUDW),并将其应用于电能质量扰动信号的检测定位。该形态小波包含了开闭与闭开组合滤波器与可检测突变信号上下边缘的形态梯度,完全满足信号重构条件。利用MATLAB对单一扰动信号与混合扰动信号进行了仿真验证,并与前人构造的形态非抽样小波进行了对比分析。结果表明,自定义的形态非抽样小波具有很好的扰动信号特征描述能力及抗噪性能,即使在强噪声环境下,也能正确指示扰动起止时刻以及突变极性。在单一扰动识别方面,论文首先提出了一种基于数学形态学与动态时间扭曲的识别算法(Mathematical Morphology-Dynamic Time Warping, MM-DTW)。算法首先利用形态滤波器对扰动信号进行滤波处理,再利用dq变换提取信号的幅值特征,最后将处理后的扰动测试信号与参考信号的特征模板进行匹配,计算信号之间的距离矩阵,采用动态时间扭曲算法在距离矩阵中寻求测试信号与各类参考信号间路径长度最短的最优扭曲路径,根据两信号距离最小、相似度最大的原则,选取路径最短的一次匹配为识别结果。仿真验证表明该算法识别效果较好,且不受原始扰动的幅值变化、扰动发生时刻(包括过零点扰动)、持续时间变化以及噪声强度的影响,具有较好的适应性。其次,论文专门针对振荡暂态与脉冲暂态两类高频暂态扰动,提出了一种基于高阶累积量(Higher order cumulants, HOC)与支持向量机(Support vector machine, SVM)的识别算法。该算法利用高阶累积量提取脉冲暂态与振荡暂态两类扰动的3阶与4阶统计特征,并选取各阶统计结果中的极大值个数、极小值个数以及最大值、最小值共8个特征量作为支持向量机的输入,从而得出识别结果。该算法在有噪声干扰与其他扰动干扰的情况下也能正确识别出两类高频扰动。在混合扰动的识别方面,论文基于小波变换、Teager能量算子(Teager Energy Operator, TEO)以及prony算法,构建了一个适用于混合电能质量扰动的识别系统,本系统首先对混合扰动进行小波变换,将其分解到不同的频率空间;再利用TEO算法对低频空间的信号分量进行分析,判断信号是否发生了频率波动、电压暂降、电压暂升或电压中断扰动;紧接着采用prony法分析去除低频分量后的其他分量,判断信号是否包含有谐波、间谐波与振荡暂态扰动;最后对高频分量进行扰动起止时刻与加窗的扰动能量计算,判断信号是否发生脉冲暂态或是电压缺口扰动。利用大量的理论混合扰动信号与PSCAD仿真信号对此识别系统进行了验证,结果表明,本系统对混合信号的识别具有较好的适应性,识别率较高。在扰动源定位方面,论文首先研究了两处电容器同时投切时的定位问题,以电容器投切造成的暂态信息为切入点,利用联网监测信息计算得出扰动能量与支路电流变化值,并将这两种特征向量输入到概率神经网络,从而得出电容器投切时的具体位置。论文还讨论了系统发生暂降扰动的定位问题,通过计算各条线路上的有功功率与无功功率的大小及变化方向,判断暂降源所在位置以及暂降原因,该方法可有效定位及辨识由电容切除、对称故障、不对称故障以及感应电机引起的电压暂降源。总的说来,本论文最终形成了一个从扰动检测到扰动识别(单一扰动与混合扰动)再到扰动定位的电能质量扰动智能分析框架。本论文是教育部留学回国人员科研启动基金—“基于联网的电能质量扰动检测、识别与定位算法及系统研究”的组成部分。

【Abstract】 Nowadays, our country is stepping into an accelerating period of industrialization and urbanization. The requirement of power supply keeps fast increasing all the while, which brings huge opportunities to the development of interconnected power grid and the scale-up of power system. However, the coming substantive power electronic equipments, impactive and nonlinear loads in power system lead to a series of power quality problems, such as voltage sag, voltage swell, voltage fluctuation, harmonics, voltage notch, impulse, oscillation transient and so on. Hence, it is significative to real-time monitor, intelligently analyze those power quality disturbances (PQD), and to build the analysis frame of them. It can provide solutions and methods for PQD detection, recognition and source location. Besides, it can supply decisions for the PQ management. Thus, the dissertation researches on the detection, recognition and source location of PQD based on the study of steady disturbances, transient disturbances and multiple disturbances.About disturbances detection, firstly, the dissertation proposes a new detection scheme on basis of generalized morphologic filter and difference-entropy. Power quality disturbances are processed by generalized morphologic filter, which can maintain their original features during the denoising process. Then, the complexity of disturbances is measured by self-defined difference-entropy to locate the starting and ending time by checking the entropy values of singular points. Secondly, a new morphological undecimated wavelet (MUDW) is defined and applied for power quality disturbances detection. The proposed MUDW composed of open-close-plus-close-open morphology filter and morphology gradient operators, which can detect the verges of disturbances, besides, the MUDW satisfies the signal reconstruct condition. Simulations are done in MATLAB and a comparison with other MUDW is given. The results show that the proposed MUDW has good depicting-characteristics ability and resisting-noise performance, also, it can direct the start-end time and variation polarity rightly even in the strong noise environment.About single disturbance recognition, on one hand, the dissertation presents a novel algorithm based on mathematical morphology (MM) and dynamic time warping (DTW). In this algorithm, firstly, morphological filter is used in disturbances filtering. Secondly, a dq transform method is used in feature extraction of those disturbances. Thirdly, it calculated distances matrix between testing disturbances and six kinds of reference disturbances, then, DTW algorithm is used to search the optimum path which needs to be shortest in every distances matrix to guarantee the testing signals and reference signals resemble most. Finally, it selectes the shortest path as classification result. Simulation results show that this novel algorithm has an excellent recognition effect, which is not influenced by the amplitude, start-end time (including zero-crossing disturbance), disturbances duration time and noise strength. On the other hand, a new recognition method based on High-order Cumulants (HOC) and Support Vector Machines (SVM) is proposed for recognizing two kinds of high-frequency disturbances: oscillation transient and impulse transient. This method utilizes HOC to extract the 3rd order and 4th order statistic features of impulse transient and oscillation transient, and selects 8 features including the amount of local maxima and local minima, the value of maximum and minimum for each cumulants as input of SVM. The proposed method is available for classifying two kinds of disturbances without influencing by strong noise and other disturbances.About multiple disturbances recognition, a classification system is constructed using wavelet transform, teager energy operators (TEO) and prony algorithm. In the First place, the system uses wavelet transform to decompose the multiple disturbances into different frequency-space. In the next place, the low-frequency component is analyzed by TEO (Teager Operators) algorithm in order to distinguish whether the signal contains frequency variation, sag, swell or interruption. Furthermore, prony algorithm is adopted to analyze the signal without low-frequency component, and it can estimate the existence of harmonics, inter-harmonics or oscillation transients. Finally, it detects the start-end instants and windowed disturbance energy of high-frequency component to identify whether it is composed of impulse or notch disturbance. A mass of single disturbances and multiple disturbances are generated for testing the proposed recognition system and the results show that the system has good adaptability and high recognition rate for those kinds of signals.About disturbance source location, on one hand, it researches on locating switched capacitors on two different buses in distribution system. It calculates the average disturbance energy and average branch current variation as features by the web-based monitoring information, and those features are put into the probabilistic neural network to get the position of capacitors. On the other hand, the dissertation focuses on the source location of sag disturbance. It presents an algorithm to determine the position and reason of sag source by computing the value and direction of active power and reactive power. Simulation results indicate that this method can locate and distinguish the sag disturbances generated by capacitor clearing, symmetrical line fault, non-symmetrical line fault and induction motor starting.From the study above, the dissertation finally forms a PQD analysis frame including disturbances detection, disturbances recognition (single disturbance and multiple disturbances) and disturbances sources location.This dissertation is supported by the Scientific Research Foundation for the Returned Overseas Chinese Scholars—’Research on the power quality disturbances detection, recognition and location methods and system based on the network-based information’

节点文献中: