【作者】 张波；

【导师】 律方成；

【作者基本信息】 华北电力大学 ， 电力系统及其自动化， 2015， 硕士

【摘要】 气体绝缘组合开关电器是高压变电站的主要设备之一,其绝缘性能的好坏直接影响着大范围电力的可靠供应,在电力系统中占有重要地位。由于GIS内部绝缘的劣化,会发生局部放电。通过检测局部放电就可以对GIS的运行状态进行评估,及时发现隐患,保障设备的可靠运行。超声波检测法利用检测局部放电发生时产生的超声波来判断是否有局部放电发生,检测灵敏度高,受现场电磁环境的影响小,已经被广泛应用于现场GIS设备的带电检测。实践已经证明,对于GIS内部不同的绝缘缺陷,在发生局部放电时产生的超声波信号各不相同。如果可以在解体前确定缺陷类型,将有助于确定缺陷发生的位置,有针对性地安排检修工作。本文首先根据GIS内部主要的六种绝缘缺陷设计了相应的绝缘缺陷模型,六种缺陷分别为:高压导体突起,地电极尖刺,悬浮金属颗粒,自由金属颗粒,绝缘子表面固定金属颗粒和绝缘内部气泡。试验时将绝缘缺陷模型放入GIS腔体内,对其加压使其发生局部放电。绝缘缺陷模型的设计可以降低试验时所加的电压,减少噪声干扰,使检测到的局部放电超声信号的特征更加明显。对局部放电超声信号的特征参数提取采用时域与频域相结合的方式,在时域上提取了均方根、方差、绝对积分平均值、峰度和偏度五个特征参数,在频域上提取了功率谱最大值、中值频率和平均功率频率三个特征参数,利用这些参数可以很好的体现不同缺陷局部放电超声信号的特点。支持向量机是建立在统计学习理论基础上,应用VC维理论和结构风险最小化原理,借助于最优化方法的一种新型机器学习方法。实际应用中,支持向量机总是属于最优的算法之一。所以,本文选用SVM作为模式识别的分类器,应用MATLAB中的LIBSVM工具箱实现该算法。识别结果表明,SVM作为局部放电超声信号的模式识别分类器时,识别率很高。本文应用小波分析的方法对采集得到的局部放电超声信号去噪,对去噪后的信号用SVM分类器进行模式识别。结果表明,与用没有去噪的信号进行模式识别相比,去噪能够提高模式识别的识别率。更多还原

【Abstract】 Gas insulated switchgear is one of the main equipments in the high voltage substation. Its insulating performance directly influences the reliability of power supply. It plays an important role in the power system. Due to deterioration of insulation in the GIS, partial discharge will occur. Through the partial discharge detection can evaluate the running status of GIS, find hidden dangers in time and ensure reliable operation of device. Ultrasonic detection method uses ultrasonic which is generated from partial discharge to determine whether the partial discharge occurs. Ultrasonic detection method has high detection sensitivity and is not affected by the electromagnetic environment. It has been widely used for GIS detection in the field. Practice has proved that ultrasonic signals generated from different insulation defects are different. If we can identify the types of defects before open GIS, this will be helpful to determine the location of defects and arrange the repair work.Firstly, according to the main six kinds of insulation defects in the GIS, the insulation defect models were designed. The six kinds of defects are high voltage conductor protrusions, spikes on ground electrode, suspended metal particles, free metal particles, fixed metal particles on the surface of the insulator and bubble in the insulator. In the experiment, the insulation defect models were put into the GIS cavity. By using the insulation defect models, the voltage used in the test was lower. This could reduce the noise interference and enable the characteristics of partial discharge more obvious.Characteristic parameter extraction of partial discharge combines the time domain and the frequency domain. In the time domain, RMS, VAR, absolute integral mean value, skewness and kurtosis are extracted. In the frequency domain, the maximum of power spectrum, median frequency and mean power frequency are extracted. These parameters can reflect the characteristics of the partial discharge ultrasonic signals that grasped in different defects very well.Support vector machine is established based on statistical learning theory. It is a new machine learning method and uses the VC dimension theory, structural risk minimization principle and optimization method. In practical application, support vector machine is always one of the best algorithms. So, in this paper, we chose SVM as the pattern recognition classifier and used LIBSVM toolbox in MATLAB to realize the algorithm. Recognition results show that when SVM was used to partial discharge pattern recognition, the recognition rate was high.The method of wavelet analysis was used to do the denoising ultrasonic signals of partial discharge. The signals after denoising were used for pattern recognition with SVM classifier. The results show that compared with original signals, denoising can improve the recognition rate of pattern recognition.更多还原