【作者】 符玲；

【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， 2010， 博士

【摘要】 电力系统故障暂态信号蕴涵了丰富的信息,充分利用这些信息,对实现电力系统故障识别、故障检测、继电保护等应用具有重要的理论和现实意义。故障暂态信号的分析和应用中,存在的一个关键问题是如何有效可靠地提取出蕴含在故障暂态信号中的故障特征信息,一方面,由于故障暂态信号通常混杂在稳态量中,而且由于能量低、幅值小,往往容易被稳态量和系统噪声所淹没；另一方面,由于故障暂态信号所包含的故障信息往往是海量的、不规则数据信息,难以用来直接达到故障识别研究的目的。这两方面的原因,就给故障信号特征信息的有效提取造成了困难。针对故障识别中特征信息提取困难的问题,尤其是对大量不确定信息的处理难题,本论文在将信息理论中相关的信息测度原理引入到电力系统故障信号特征信息的提取和处理中作出了尝试和研究,并取得了较好的实验效果,从而为基于故障暂态信号的电力系统故障识别研究提供了一条新的解决思路和辅助方法。论文在研究信息测度理论的基础上,结合时域、频域和时频域多种信号变换空间内的信号处理方式,提出了推广和泛化的信息测度概念。这是因为信息测度理论本身是定义在时域范围内的测度指标,而电力系统故障信号的特征信息往往需要从不同的信号变换空间来综合研究,因此将信息测度理论推广到不同的信号变换空间会更有利于对故障信号特征信息进行处理。文中主要研究了信息熵测度指标和近似熵测度指标在对故障特征信息进行处理中的应用,具体是将时域信息熵、频域信息熵、小波信息熵以及时域近似熵、时频域近似熵这5种信息测度应用到励磁涌流和故障电流识别、高压输电线路故障类型识别、小电流接地系统中故障线路识别、低频振荡中故障信号识别以及超高压输电线路故障类型识别中。论文首先研究了时域信息熵测度在变压器励磁涌流和故障电流识别中的应用。时域信息熵是信息熵测度与时域统计分析方法相结合的产物,基于时域信息熵的励磁涌流和故障电流识别算法在非对称性励磁涌流、对称性励磁涌流、电流互感器饱和、合闸到故障以及受到较大谐波干扰等多种工况下都能正确地识别励磁涌流和故障电流,并且克服了传统识别方法中阈值设定、互感器饱和影响、数据窗的选取、波形的拟合等固有局限,体现了时域信息熵在特征信息处理中的优势。其次,论文研究了小波信息熵测度中的小波奇异熵在高压输电线路故障类型识别中的应用。小波奇异熵是一种基于时频域分析的信息熵测度,是小波分析和信息熵相结合的产物。文中将小波奇异熵应用于高压输电线路故障类型的识别,并在不同的故障类型、故障电压初始角、故障过渡电阻和故障位置等工况下仿真对故障类型识别的效果,仿真结果证明基于小波奇异熵的故障识别算法基本不受到故障工况和系统噪声的影响,而且算法计算复杂度也满足应用的需求。验证了小波信息熵算法在故障类型识别中取得了理想的实验效果。再次,论文研究了近似熵测度对理想电力信号内部复杂性的分析结果,并将近似熵测度和小波变换相结合的时频近似熵测度应用到小电流接地系统故障选线和低频振荡下故障信号识别中,通过对故障线路以及不同类型故障的正确识别,再次验证了信息测度在故障识别中的可行性和有效性。最后,论文在研究信息融合思想的基础上,建立了融合多种信息测度的特征信息来研究故障识别的模型,一方面,建立了多种小波熵测度证据融合模型,将来自四种不同小波熵测度的特征信息进行融合；另一方面,建立了基于“频域信息熵测度-时域近似熵测度”的复合信息测度模型,将不同信息测度的特征信息进行融合。信息测度的融合克服了电力系统故障识别中诸多不确定因素的影响,通过第一种信息融合模型在小电流接地系统的故障线路识别和故障类型识别中的应用,以及第二种信息融合模型在超高压输电线路故障类型识别中的应用,验证了信息融合模型的正确有效性。更多还原

【Abstract】 The fault transient signal in the power system contains rich information, and making use of this information can help to realize the power fault identification, fault detection and relay protection. During the analysis and application of the fault transient, there would be a critical problem to be solved which is how to extract the fault feature information from the fault transient signals effectively and reliably. On one hand, because of its intermixture with the stable signals and due to its low energy and magnitude, the fault transient signals are tending to be affected by the stable signals and the systematic noise. On the other hand, the information data contained in the fault transient signals is usually enormous and irregular, so it is difficult to employ the information directly to reach the fault discrimination purpose. Thus, these two aspects of reasons make it a tough issue to extract the feature information effectively from the fault transients. Aiming at the extraction issue of fault feature information in the fault discrimination, especially the processing issue of the enormous and uncertain information, this thesis tries to introduce the information-measurement theory into the fault feature extraction. The information-measurement theory is good at processing the information and has been applied successfully in other fault research fields, and its application in this thesis has been proved to be valid as well through the testing. Therefore, the information-measurement based method for feature extraction provides a novel solution or assistant way to the fault discrimination in power system.Based on the study of information-measurement theory, this thesis combines it with the signal-procession methods of different domains, like time domain, frequency domain and time-frequency domain, and proposed the concept of extended information-measurement. This is because the information-measurement theory defines the measurements originally in the time domain only, but the feature information of the power fault should be usually considered in various domains. Therefore, the extension of the original information-measurement theory into different domains can be beneficial to the processing of the fault feature information. The thesis mainly discusses the application of the information-entropy measurement and the approximate-entropy measurement into the fault feature extraction. The research has used the time-domain information-entropy, frequency-domain information-entropy, wavelet entropy and time-domain approximate entropy, time-frequency-domain approximate entropy into the application of fault identification.First, this thesis studies the application of time-domain information entropy into fault discrimination. The time-domain information entropy is the combination of information entropy and the time-domain statistic analyzing method, and it is applied to discriminate the magnetizing inrush from the fault current. The simulations under various conditions, like dissymmetrical inrush current, symmetrical inrush current, CT saturation, inrush current switching onto fault, prove that this method can overcome the limitation of the traditional method and verifies the advantage of the time-domain information entropy in information processing.Second, the thesis studies the application of wavelet singular entropy into the fault-type discrimination. The wavelet singular entropy is one kind of the time-frequency domain information entropy which combines wavelet transformation with information entropy. The thesis employs it into the fault-type discrimination in high-voltage transmission line uner various conditions, like different fault type, fault inception time, fault resistance, fault position, and has obtained a satisfactory result.Third, the thesis studies the analysis of ideal power signals by the approximate entropy, then studies the application of time-domain approximate entropy into the fault-line selection and fault identification during power swing. The simulation results verify the feasibility and validity of the information-measurement based fault discrimination.Finally, according to the information fusion theory, this thesis fuses the feature information from various information measurements together to discriminate the faults. On one hand, we establish the fusing model of different wavelet entropies to fuse the information from 4 different kinds of wavelet entropy measurements; on the other hand, we establish the’frequency-domain information entropy and time-domain approximate entropy’ model to fuse the information from different measurements. The first model has been used into the fault line and fault type discrimination, and the second model has been used into the fault type discrimination in EHV transmission line. Through the fusion of information measurements, the effect of systematic uncertainty in the fault study can be overcome and a better fault discrimination can be achieved.This thesis is supported by National Natural Science Foundation of China:’Wavelet entropy theory and its application in power fault detection and classification’(No.50407009, 2005-2007) and’Information theory based power network fault diagnosis of multi-sourced signal’(No.50877068,2009-2011).更多还原