【作者】 乐全明；

【导师】 郁惟镛；

【作者基本信息】 上海交通大学 ， 电力系统及其自动化， 2006， 博士

【摘要】 电网的安全运行越来越依赖于对各种信息的有效分析和处理。电力系统发生故障,尤其是大面积复杂故障后,仅依靠来自SCADA系统和微机保护间隔单元的保护、开关接点的变位数字量信息,调度员难以做出准确判断。而来自继电保护和故障录波器的信息越来越成为事故分析和系统恢复的重要依据。对这些信息进行分析,可以得出系统异常时的故障类型、故障位置、故障相别、故障前后的相量、保护及开关动作事件、事件时序(SOE)等;对来自多个厂站的保护和录波信息进行综合分析,可使调度中心迅速掌握电网真实故障情况,以及继电保护、开关和重合闸的动作行为,这对调度员迅速做出正确判断并恢复系统供电创造了条件,同时有助于继电保护人员和调度员利用故障数据进行事后分析。本文充分利用高压/超高压电网中故障录波器记录的原始海量故障录波数据,及录波数据中的暂态信号,借助新的数学方法如小波、神经元网络、时态逻辑技术等,对传统电网故障诊断中的系列算法进行了改进和创新,取得了以下主要成果:◆提出了基于信号奇异性检测原理的故障录波数据自适应小波去噪压缩新算法,详细分析各尺度上小波系数奇异点的匹配搜索过程和最大尺度层数自适应阈值选取方法;通过故障录波信号频率与采样率的关系,来确定最大小波分解层数n,然后在该分解层数上进行噪声白化检验,并对某一个信号奇异点进行奇异性指数计算,以确定出小波最优分解层数;论文还系统阐述了基于小波模极大值的信号重构算法。通过大量真实故障录波数据编程仿真,验证了算法的高效性。◆在子站端,提出了基于二进小波和信号奇异性检测原理的数据预处理和故障初步诊断新方案,通过二进小波准确提取电网故障发生时刻,以解决录波器启动记录时标存在的误差问题,故障时刻提取误差率不超过0.6ms。同时,提出了一种基于小波理论的故障选线和选相新思想,为全网实时或准实时故障诊断创造了条件。◆在主站端,提出了基于小波神经网络和故障电流量的故障类型识别新算法,利用bior3.1提升小波和RBPNN网络构造了新的小波神经网络故障类型识别模型,应用bior3.1提升小波对故障前一个周波和故障后两个周波的电流量进行分解,将分解到的(0～375)Hz频率段的小波系数输入到神经网络。通过ATP仿真测试,选线准确率达100%,证明了该模型的高效性。◆提出一种新的系统振荡与故障信号识别方法:利用双正交小波包和连续复小波,分别提取故障信号小波系数局部模极大值、不同频段小波系数变化率、电流小波系数幅值变化趋势以及电流与电压小波系数相角差作为新判据,根据各判据的特点与适用性,运用模糊集合理论,分别给予各判据不同的隶属度,最后给出振荡与故障识别综合判据,大量仿真验证了该算法的有效性。◆提出了基于时态逻辑技术的全网故障诊断新思想,其核心是利用全网故障线路及与故障线路相关的模拟量信息准确提取和校核开关量信息。利用线性时态逻辑技术,建立了故障诊断演绎模型,根据全网保护配置信息和配合关系,利用模拟量信息分层和保护分级诊断思想,诊断出保护、开关和重合闸动作行为,形成故障简报,供调度员参考决策。通过仿真测试,故障准确识别率较高,基本解决了传统基于数字量信息故障诊断存在的缺陷。为验证各功能模块集成运行性能,运用MATLAB的GUI编程功能,开发了故障信息处理系统仿真验证系统软件,即DEMO测试系统。该系统采用分模块的形式验证了系统各算法的有效性,并实现软件的可视化和自动化。同时,为了验证基于时态逻辑技术的主站故障诊断模块,本文基于CLIPS平台建立了电网故障诊断模型,实现故障诊断推理智能化。更多还原

【Abstract】 Whether a power network would run safely depends on valid analysis and transaction of various faulted information. It is very difficult to determine the fault course for dispatchers only based on the numeric information of protections and breakers. But the faulted recorder data will be the important basis of analyzing and diagnosis of power system more and more. The fault types, location, phase faulted, phasor of electricity before and after fault, action of protection and breakers, sequence of events can be determined by analyzing faulted recorder information. So the dispatchers can seized the real situations of the fault, and determine how to supply power again according to all the faulted record data that come from various substations. In the same time those information can be analyzed by protection personnel and dispatchers after a fault happened.The all original faulted recorder data have been used in this paper. And the new math tools, i.e. wavelet, NN, temporal logical technology and so on are introduced. The traditional algorithms of fault diagnosis have been improved and innovated in the paper. The main achievements will be concluded as follow. The reasons to compress and transmit the fault-recorded data in HV substations in time are expanded. The virtues and disadvantages about the traditional data compression methods are analyzed. The novel eliminating noise self-adaptive data compression method based inspecting signal singularity is set forth in the first time in this paper. The matching search process of wavelet coefficient singularity in every wavelet scale and the self-adaptive threshold method of the maximum scale number are analyzed. The maximum decomposing scale n is determined by the sampling rate and frequency of faulted record data, and then inspecting the white noise and calculating the singularity exponent on the n scale can determine the optimum decomposing scale. The signal reconstruction method is discussed in the last. The validity is testified by some real faulted record data from East China power network.An advance data preprocessing and diagnosis scheme in substations is put forward according to dyadic wavelet and singularity theory. Dyadic wavelet can locate the fault time point, which could solve the time error of faulted recorders. And its error is smaller than 0.6ms after they are revised. A new algorithm of selection line and phase is presented based on wavelet theory, which brings a good condition for real-time fault diagnosis.In order to improve fault recognition capability and computational speed of the fault diagnosis system, this paper presents a new wavelet neural network mode constructed from lifting wavelet and PNN neural network. The coefficients of fault currents in the low frequency band between 0 and 375 Hz that decomposed by bior3.1 lifting wavelet are put into the neural network. Through ATP simulation and the test of real fault record data from the power network in East of China, the result indicates that the mentioned model in this paper has very high recognition rate and convergence speed.A new scheme to distinguish oscillation and fault occurring in transmission lines is proposed based on wavelet transform. So a new synthesized criterion consists of four methods as following: the first, comparison of the local maximum wavelet coefficients of the singularity point; the second, how those wavelet coefficients mentioned above changes in different frequency band at the very moment; the third, how those wavelet coefficients changes as time goes by in the same frequency band; the fourth, how the phase between voltage and current changes as time goes by in the same frequency band. Each method is evaluated by the performance and applicability based on fuzzy set theory. The algorithm proves to be feasible and of high veracity in typical experiments.A new idea was presented in this paper, which is about Linear Temporal Logic (LTL) technology and analog information be used in fault diagnosis system for High Voltage (HV) transmission lines. A linear temporal logic formal deduction system is expressed with syntax and semantic. Fault analog signal is preprocessed in substation, considering need for system fault tolerant and real-time requirement of fault diagnosis. In dispatch center, time sequence of protection and breaker actions in fault diagnosis process is analyzed, then, TDS and TAS about typical fault mode are formed and time sequence restriction relation is expressed by LTL. The reliability and tolerance of reasoning in this paper is verified by instance at last.The DEMO system programmed by MATLAB GUI can validates the integration system consisted by every function module. And the temporal logic diagnosis of protections and breakers is proved by CLIPS simulation tool. That validates the effective and intelligent diagnosis system.更多还原