【作者】 万元；

【导师】 李朝晖；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2009， 博士

【摘要】 大型发电机作为系统内的主电源，是电力系统的关键设备，其运行的安全性历来受到高度地重视。发电机安全运行的主要威胁之一来自绝缘体系，而局部放电活动是定子绝缘劣化的重要征兆。因此，长期以来，局部放电在线监测已被应用为检测和诊断发电机绝缘状况的重要手段，是发电机状态监测的重要内容，也是人们研究的热点与难点。在对当前发电机局部放电在线监测方法及研究现状进行总结的基础上，结合葛洲坝电厂的具体需求及其发电机的具体参数，通过理论与实践相结合，开展了发电机局部放电在线监测与分析方法的研究。针对葛洲坝电厂监测需求，设计了一套大型发电机中性点局部放电在线监测系统，应用中性点耦合法获取局部放电信号，采用分层式网络结构，为用户提供局部放电集成分析与远程诊断平台。论证了局部放电传感器在获取高频信号时的有效性。分析了系统采用的中性点耦合法与目前流行的PDA法相比，在设备成本、可靠性、对中压绕组局部放电的灵敏度、监测结果全面性等方面存在的优势。应用高速采集与分析技术，增强了系统对定子绝缘瞬态故障的反应能力。为提高监测的灵敏度，设计了放大倍数切换方法。鉴于机组运行状况对局部放电的影响，为了在减少冗余度条件下捕获全面的局部放电状态信息，设计了多种数据存储机制。信息存储中同时考虑机组运行工况的变化，机组负荷，信号的突变，信号的大小等因素。结合葛洲坝电厂发电机的具体参数，全面系统地研究了局部放电传感器的可靠性问题，包括局部放电传感器本体的可靠性及其安装的可靠性。论证了发电机正常运行状况下局部放电传感器本体的可靠性。采用稳态分析的方法，计算了发电机出口母线处出现不对称短路故障时中性点稳态电压的大小，验证了发电机异常状态下局部放电传感器本体可靠性。研究了局部放电传感器的安装对发电机单相接地故障时的零序电压保护、故障接地电流、三次谐波电压保护三个方面及高压系统的传递过电压的影响。应用暂态网络分析模型和准分布参数模型仿真了在发电机定子出口母线及内部出现单相接地故障时，局部放电传感器的安装对健全相及中性点暂态电压的影响。理论上分析了背景噪声的时域能量分布特征，设计了确定脉冲边沿的滑动时窗能量法。在有效提取单脉冲波形的基础上，提出了两种脉冲识别方法分别为基于自适应聚类及三维谱图的脉冲识别方法和基于时频联合特征提取的脉冲识别方法。总结了自适应聚类算法在实际工程应用中存在的问题，并提出了有效的解决方案。提出了基于小波能量分布和基于频域空间能量分布的脉冲频域特征提取方法，针对小波分析中存在的分解层数难于确定、小波基函数难于选择、消噪阈值难于选取三方面问题，分别提出了解决方案：根据脉冲的最低识别频率确定小波分解层数、应用相关系数法确定小波基函数、采用基于3σ的阈值方案。将二维图形的统计参数应用于局部放电脉冲识别。在介绍两种方法的基础上，总结了两种方法的优点和缺点，最终确定基于时频联合特征提取的脉冲识别方法作为现场应用的最佳方案。为了减少局部放电在线监测中的信息浪费，设计了脉冲综合分析方法，对不同激励源的脉冲实施相应的专项分析，以诊断其激励源的状态，最终实现发电机综合运行状况的监测与诊断。设计了局部放电脉冲、电弧脉冲、励磁系统脉冲等的分析方法。针对传统发电机局部放电在线监测系统的不足，提出了局部放电的集成分析方法，融合发电机的运行状态进行局部放电分析，保证分析结果更加准确。大型发电机中性点局部放电在线监测系统已经在葛洲坝电厂获得应用，获取了丰富的脉冲信号，基本上建立现场脉冲的知识库，监测和分析结果能较好地体现了定子绝缘的状态，为优化发电机运行、指导发电机状态检修提供了有力的依据，对于促进局部放电在线监测研究有着十分重要的现实意义。更多还原

【Abstract】 Large generators, as the main power source, are the key equipment in the power system, so the safety of their operation has always been the highest priority. One of the main threats of generator’s safe and reliable operation is from insulation system. The partial discharge (PD) activity is an important symptom of stator insulation deterioration. Therefore, the on-line PD monitoring has been used as an effective method for a long time to detect and diagnose insulation deterioration of generator, which is an important content of generator condition monitoring and also a hot and difficult study. Based on summarizing the existing methods and researches on PD monitoring technique, this dissertation is focused on research the generator on-line monitoring and analysis method, the research is combined with the needs of Gezhouba Hydro Power Plant and the specific parameters of its generators. The methodology and technique are comprehensively investigated on integration of theory with practice.In order to meet the needs of Gezhouba Hydro Power Plant, a PD on-line monitoring system mounted at neutral for large electric generators is designed, which uses the wideband measurement technique to obtain PD signal in the generator’s neutral point. To provide a PD integrated monitoring and remote diagnostics platform, the system adopts the framework of hierarchical network structure. Then the effectiveness to obtain high-frequency signals of the PD sensor is demonstrated. And the comparison between the method of PD sensor mounted at neutral and the PDA method which is popular is carried out, it is found that the method of PD sensor mounted at neutral has the advantage in the equipment cost, reliability, comprehensive monitoring results and sensitivity of detecting PD occurred at the medium voltage windings. The high-speed data collection and analysis techniques are used to enhance the response capability to the stator insulation transient failure. Aiming at enhancing the sensitivity of the monitoring, the automatic amplification switching method is designed. Particularly, considering the influence of the operating states of the generator on PD, a variety of storage mechanisms for data storage is proposed to capture all PD features in different operating states with less data redundancy, and the operating conditions, load, signal mutation, signal strength and so on are all took into account in data storage.Employing the specific parameters of generator in Gezhouba Hydro Power Plant, the reliability issues of the PD sensor was systematically analyzed, including the reliability of the PD sensor body and the the reliability of the installation of PD sensor. The reliability of PD sensor body during normal operation of generator is explained firstly. By using the steady-state computational method, the magnitude of the neutral steady-state voltage was investigated when there was an asymmetry short fault at the generator expert bus, and then the high reliability of PD sensor when the generator in asymmetry short fault is verified. Subsequently, the impact of the PD sensor installation on the generator’s single-phase grounding fault including zero sequence voltage protection, third harmonic protection, grounding fault current and high-voltage system transmission overvoltage is investigated. Two simulation models, which are transient network analysis model and quasi-distributed parameter model, are separately adopted to investigate the impact of the PD sensor installation on the overvoltage of non-fault phase and the neutral when the single phase grounding fault occurs at export bus and internal stator windings of the generator.The time domain energy distribution character of the background noise is studied in theory, and the sliding energy window method which could determine pulses boundaries is designed. Based on the effective extraction of the single pulse waveform, two pulses identification methods (a) adaptive clustering algorithm and three-dimensional distribution (b) synthetic features extraction in time and frequency domain are proposed. Some effective schemes are proposed to solve out the problem of the adaptive clustering algorithm when used in practical application. Wavelet energy distribution and spectra space energy distribution are extracted as the pulses frequency domain features. To overcome the three problems in wavelet analysis, which are difficult to determine the decomposition level, difficult to select the mother wavelet and difficult to choose the denoising threshold, solutions, including adopting the lowest identification frequency of pulses to determine the decomposition level, applying the correlation coefficient method to select the mother wavelet and adopting the 3a threshold method, are put forward respectively. The statistical parameters of two-dimensional diagram were applied for pulses identification. The basic principles of both pulse identification methods are introduced and the comparison between both methods has been undertaken. Finally, method (b) was determined as the best option for field application.The pulses comprehensive analyzed method is designed to decrease the information waste for on-line PD monitoring, which implements corresponding special analysis to different excitation pulses, diagnoses the state of the excitation source, and finally achieves the monitoring and diagnosis of the generator comprehensive operation condition. Methods to analyze PD pulses, arc pulses and pulses generated by excitation system have been respectively set up. Traditional monitoring system only analyzes PD in isolation, which is unreasonable. Based on investigating the influence factors of PD, the PD integrated analysis method is employed to ensure a more accurate result, which combines with the operation state information of the generator to implement PD synthetical analysis.Nowadays, the PD on-line monitoring system mounted at neutral for large electric generators has been put into on-line operation in Gezhouba Hydro Power Plant, which has got plenty of pulses signal and basically established the knowledge database for filed pulses, it’smonitoring and analysis results can describe the stator insulation state well. All the achievements are meaningful for carrying out condition-based maintenance and improving PD on-line monitoring research for generator.更多还原