【作者】 董越；

【导师】 肖登明；

【作者基本信息】 上海交通大学 ， 电气工程， 2008， 博士

【摘要】 实践证明,断路器状态监测为实现整个电网的安全运行提供了有力的保障,同时对保证高压断路器的可靠运行、减少因计划检修带来的巨大人力、物力资源的浪费等,具有极大的应用价值。针对目前断路器在线监测系统存在的功能单一、缺乏有效检测手段等问题,本课题组与上海电力公司联合开展了SF6高压断路器综合在线监测系统的开发。根据SF6断路器的结构特点及分/合闸的动作过程,针对动触头行程、机械振动信号,分别设计了滑阻型位移传感器及压电式加速度传感器的安装;根据该类型断路器的具体运行参数,分别给出了求解断路器分/合闸平均速度的修正算法;具体分析了周围环境对SF6气体微水含量的影响,结合SF6断路器气体状态参量的正常工作范围,认为周围环境对SF6断路器中的SF6气体微水含量的影响很小。对现场信号进行小波包频带能量分析发现,反映断路器机械状态变化最为敏感的频带分布比较分散,没有规律地集中在某一或几个频带上,只能选择多组实验结果中出现概率较大的频带作为振动信号的特征频带。根据现场数据及文献的实验结果进行公式推导证实,正常工作状态或状态变化较小的情况下,断路器的振动信号可近似看成一平稳随机过程。采用小波包频带能量分析结合AR模型(Auto-Regressive Model)功率谱的方法,对断路器振动信号进行状态检测发现,对于现场采集的三组机械状态相近的断路器振动信号,其各自的AR模型功率谱没有呈现相似性,该方法不能作为断路器振动信号状态检测的研究方法。针对方法失效的问题,分析了现场噪声的影响,认为噪声对功率谱估计的影响不大,不应是导致方法失效的主要原因;功率谱估计的前提是必须假设信号满足线性和高斯分布,断路器振动信号的实际特性与之是否匹配,才是失效的主要原因。首次采用小波包频带能量分析,结合双谱估计的方法,对断路器振动信号进行状态检测。现场实例分析表明,断路器振动信号为非高斯分布、非线性特征的信号,这与双谱方法的前提条件相符,与功率谱方法的假设前提条件相悖;对于几组状态相近的断路器振动信号,采用非参数化方法中的间接法,其结果能呈现基本的相似性,但效果仍无法让人满意;采用参数化模型方法的结果证实,无论采用ARMA模型(Auto-Regressive Moving Average Model)或AR模型,其效果都好于间接法,几组信号的特征谱图呈良好的相似性;但短数据(特征频带信号)时,ARMA模型中MA(Moving Average)部分的估计误差较大,相似的程度没有AR模型的好;而适用于短数据处理的AR模型,则表现出其优越性,得出的几组结果非常相似;同时,对于状态变化较大的信号,其谱图变化也很大,具有明显的可分性,该方法适用于断路器振动信号的状态检测。最后利用功率谱理论上抑制白噪声、双谱理论上完全抑制高斯噪声的特点,对现场噪声进行分析,结果表明现场的断路器振动信号含有较多的非高斯噪声,且在断路器振动信号的特征频带部分(高频部分),存在较多的非高斯白噪声;振动信号的非线性特征几乎存在于整个频率范围,且主要集中在特征频带部分(高频部分),这间接说明断路器振动信号的特征频带部分包含了绝大部分反映断路器动作过程中机械状态变化的有用信息,其提取出的特征参量能够表征断路器的机械状态的变化情况。更多还原

【Abstract】 Practice shows that development of CB (Circuit Breaker) condition monitoring system provides a powerful guarantee on security of power network operation, which is valuable for state detection of circuit breaker in reliable operation of high voltage CB and decreases the cost of human and material resources for planned overhaul. Due to the simplicity of system function and lack in effect state detection etc., an overall SF6 high voltage CB on-line monitoring system is developed by our team and Shanghai electric power corporation.According to the structure of SF6 CB and its operating characteristic, the installation of slip resistance type displacement sensor and piezoelectric acceleration sensor are designed for acquisition of mobile contact travel and vibration signal of CB respectively. The computation of average velocity of CB mobile contact is modified with the detailed operation parameter of CB. The effect of surrounding environment on SF6 moisture content is analyzed, which shows that the effect is not heavy as it was worried about with reference to the operation range of CB SF6 gas state parameters.With analyzing CB field vibration signal by WP (Wavelet Packet) band energy analysis method, the band sensitive to state change most is not regularly distributed on one band or several bands, thus the bands with relative high occurring probability can be chosen as the feature band. According to formula derivation of field data and test results of cited documents, it is proven that CB vibration signal can be considered as stationary stochastic process in normal operation condition or small changes of CB operation condition.It is testified that the method that combining WP band energy and AR model (Auto-Regressive model) PSD (Power Spectrum Density) is not fit for detecting the condition of CB vibration signal because the AR model PSD of the three vibration signals which mechanical work state change is very small is not similar. The effect of field noises is analyzed for the unfitness of the method mentioned above, which can be concluded that the effect of noises on PSD is not the main reason for the mthod unfitness. It is the true reason that the premise of Gaussian linear signal does not match the characteristic of CB vibration signal.The method that combining WP band energy method and bi-spectrum estimation is presented for the condition detection of CB vibration signal in this paper. Although the analysis of field case shows that indirect method used for bi-spectrum estimation on CB feature band signal has certain validity, the analysis result still can not satisfy the actual need. It is testified that either ARMA model (Auto-Regressive Moving Average model) method or AR model method is superior to indirect method. The AR model method which is good at processing the short length signal is more fit for the condition detection of CB vibration signal than ARMA model method, because the relative big estimation error in MA (Moving Average) part of ARMA model wakens the superiority in estimating short length signal.Based on the characteristic of PSD restraining white noise and bi-spectrum restraining Gaussian noise, the PSD method and bi-spectrum estimation is used to analyze the composition of field noise, which show that there is much of non-Gaussian noise in CB field signal and there is much of non-Gaussian white noise concentrated on CB feature band. The conclusion that the nonlinear characteristic exists in almost all the frequency band of CB vibration signal and most of the nonlinear characteristic is in the CB feature band, testifies that the CB feature band includes most of informations sensitive to CB mechanical state change and the extracted characteristic parameter can reflect the changes of CB mechanical state actually.更多还原