【作者】 李佐胜；

【导师】 姚建刚；

【作者基本信息】 湖南大学 ， 电气工程， 2009， 博士

【摘要】 随着经济的平稳快速发展,一方面电力系统输电线路电压等级不断提高、电网规模不断扩大,另一方面大气污染加剧。暴露在污秽条件下的绝缘子表面会沉积污秽,当遇有大雾、毛毛雨、融雪等不利气象,易造成电网污闪事故。污闪已经成为电网安全最具危害的影响因素之一,严重威胁着供电的可靠性。实现绝缘子污秽程度的安全、准确监测,使输电线路由计划检修向状态检修转变,是输电线路安全运行迫切需要解决的问题,对于污闪问题的解决具有重要意义。红外成像可实现绝缘子污秽等级的非接触性测量,不易受电磁干扰,安全、经济、便捷。本文系统地对运行绝缘子发热理论、图像去噪、图像分割、盘面图像提取、污秽特征提取、污秽等级分类、成像视角、特征选择等关键技术问题展开深入研究,具体工作有以下几个方面:1.红外热像检测绝缘子污秽等级缺乏完善的发热理论支持。针对现有绝缘子发热模型不能处理表面出现干燥带及干燥带电弧的问题,基于绝缘子表面水分蒸发主要取决于表面发热的假设,提出了一种湿污绝缘子表面发热分析方法,引入污层表面电阻率、湿润强度、电弧模型,建立了干燥带及干燥带电弧产生的判断条件和不同运行状态绝缘子的发热模型,并应用数值分析方法进行求解。计算机仿真结果揭示了不同运行状态绝缘子的表面发热分布规律、出现干燥带及干燥带电弧对泄漏电流和发热的影响。湿污绝缘子的红外热成像试验结果表明,该模型合理,为绝缘子污秽等级红外热像检测提供理论支持。2.绝缘子红外热像对比度低、噪声大,必须采取有效措施准确恢复绝缘子表面温度场信息。首次证实了绝缘子红外热像小波系数服从拉普拉斯分布,利用平稳小波变换分解系数冗余有利于处理具有统计规律的图像,提出了一种基于最大后验(MAP)估计的平稳小波域局部自适应绝缘子红外热像去噪方法,使用待估计点圆形邻域系数估计信号方差,并根据图像噪信比自适应调整邻域窗口大小,采用MAP估计器对各高频子带小波系数进行局部自适应估计,最后对处理后的小波系数进行平稳小波反变换得到去噪后图像;利用双树复小波变换具有近似的平移不变性和良好的方向选择性的优点,提出了一种基于MAP估计的复小波域局部自适应绝缘子红外热像去噪方法,对不同滤波器组采用各自最精细分解层子带系数估计噪声方差,利用待估计点圆形邻域系数估计信号方差,且随分辨率变化调整圆形邻域半径,使得MAP估计的无噪声系数更为准确,提高了去噪图像质量。实验结果表明,与传统的维纳滤波法、基于小波变换、平稳小波变换和双树复小波变换的贝叶斯阈值去噪方法比较,这两种方法具有更高的信噪比,在有效去除图像噪声的同时,图像细节信息保留更完好。3.实际污秽检测以单片绝缘子作为分析对象,根据截取单片绝缘子红外热像灰度直方图的特点,提出了直方图包络线分割阈值提取方法和对数变换域最大类间方差法分割阈值提取方法。以阈值提取方法为基础,提出了阈值分割与形态学后处理相结合的绝缘子红外热像分割方法。实验结果证明,运用所提方法分割后的绝缘子图像完整,边缘清晰,分割质量良好。4.绝缘子串热像相互重叠,研究感兴趣的区域为绝缘子半盘面区域,能否完整有效的从图像中提取出来,直接关系到后续污秽特征提取的有效性。绝缘子盘面图像具有椭圆特征,提出了以分割图像重心坐标为起点的不同角度散射直线来采样绝缘子盘面边缘点,应用最小二乘拟合盘面边缘椭圆方程,提取椭圆内长轴以上图像区域,获得研究感兴趣的绝缘子半盘面区域。实验结果证明,采用此方法获得了统一规范的绝缘子半盘面区域。5.红外成像设备自身存在测温误差,导致红外成像测量温度与真实温度之间的温度偏差不可预测。为了避免测量温度的误差影响,充分利用红外成像的测量精度,使温度场信息更准确、更可靠,提出了基于相对温度的污秽特征提取方法。根据绝缘子表面发热分布规律提取污秽特征,污秽等级识别综合考虑环境湿度的影响。利用整体温度分布差异,提取相对温度的平均值、方差、偏度、峭度、能量和熵6个统计参数作为污秽特征,设计了绝缘子污秽等级BP神经网络分类器;利用盘面温度随盘径变化的差异,提取径向相对温度均值作为污秽特征,设计了最近邻湿度条件下的最小距离分类器评定绝缘子污秽等级;利用热像灰度直方图间接体现相对温度分布,提取规格化灰度直方图作为污秽特征,设计了最近邻湿度条件下的灰色综合关联度最大相似准则评定现场污秽度等级。实验结果验证了这三种方法的可行性和有效性。6.确定最佳成像角度有利于提高红外热像检测绝缘子污秽等级的准确性,提出了采用Fisher准则对不同成像角度提取的相同污秽特征进行对比分析确定最佳成像角度的方法。实验结果表明成像角度变化显著的改变所得绝缘子表面热场,下盘面特征比上盘面特征有更好的分类性能。推荐红外热像检测绝缘子污秽等级应以下盘面为准。7.为了获取分类性能优异的污秽特征和较低的特征维数,提出了基于单因素方差分析的污秽特征选择方法。实验结果表明所提方法简单、有效,不但降低了数据处理的复杂性,而且避免了不良特征进入分类特征集,提高了污秽等级分类的准确性。综上所述,本文解决了红外热像检测绝缘子串污秽等级的关键技术问题,能够实现绝缘子串污秽等级的红外热像准确检测。更多还原

【Abstract】 With the stable and rapid development of economy, on the one hand, the voltage rank is getting increasingly higher and the power system scale is becoming continuously larger; on the other hand, the environmental pollution becomes severer. Pollutants are accumulated on surfaces of insulators for their exposure to the contaminant condition. Under adverse weather conditions, such as heavy fog, drizzle, snow melt, pollution flashover is easily caused within the power grid. Pollution flashover has become one of the most harmful influencing factors on the safety of the power grid. It seriously affects the reliability of the power supply. Realizing the safe and accurate monitoring of insulator pollution severity could enable the transmission lines to change from planned maintenance into condition-based maintenance. It is urgent to solve the secure operation of transmission lines, and is significant to resolve the problem of flashover. Infrared imaging can achieve a non-contact detection of insulator pollution level with many merits, such as safety, thrifty, facility and immunization to electromagnetic interference. Key technical problems—such as heating theory for running insulator, image de-noising, image segmentation, disc image extraction, pollution feature extraction, pollution level classification, visual angle and feature selection—are discussed deeply and systemically in this dissertation. The concrete works are as follows:1. It was a lack of perfect heating theory to support pollution level detection of high voltage insulators using infrared thermal imaging. As the existing heating models of insulators could not handle the problem of dry band or dry band arc on the insulator surface, a heating analytical method of polluted and wetted insulators is proposed on the assumption that water evaporation mainly depends on heat generation on the insulator surface. By introducing contamination layer surface resistivity, humid intensity and arc model, the judging condition of the generation of dry band or dry band arc and the heating model for each running state are developed and solved by numerical analysis method. The simulation results reveal the thermal distribution on the insulator surface and the impact of the dry band or the dry band arc on leakage current and heating. Infrared thermal imaging experiment results of polluted and wetted insulators indicate that the proposed model is reasonable and can give theoretical support to insulator pollution level detection by infrared thermal imaging.2. The insulator infrared thermal image is characteristic of low contrast and big noise, so effective measures must be taken to restore the real temperature distribution on the insulator surface. It is confirmed for the first time that the wavelet transform coefficients of insulator infrared thermal image obey Laplacian distribution. Because the redundancy of stationary wavelet transform coefficients is beneficial to handle the image with the statistical law, a stationary wavelet-domain local adaptive de-noising method for insulator infrared thermal image based on maximum a posteriori (MAP) estimation is developed. The noise variance is estimated using the finest scaling sub-band coefficients. The pointwise signal variance is computed with its circular neighbouring coefficients, and the neighborhood size is adjusted based on the noise-to-signal ratio of image. MAP estimator is adopted to estimate different scaling clean coefficients locally and adaptively. Finally, inverse SWT is applied to gain the de-noised image. Taking the advantage of both approximate shift invariance and good directional selectivity of dual tree complex wavelet transform (DT-CWT), a complex wavelet-domain local adaptive de-noising method for insulator infrared thermal image based on MAP estimation is developed. The author utilizes the finest scaling sub-band coefficients of different filter banks to estimate their respective noise variances, and computes the signal variance of a coefficient using neighboring coefficients within a circular window whose radius varies with resolution, so noise-free coefficients are more accurately estimated by MAP estimation and the quality of the de-noised image is improved. Experimental results demonstrate that the developed methods get higher signal-to-noise rate (SNR), de-noise more effectively and preserve more detail information of the original image than traditional Wiener filtering method,the adaptive Bayesian threshold methods based on wavelet transform, SWT and DT-CWT.3. A single insulator is regarded as analytical object in actual pollution detection. According to the characteristics of the gray histogram of the intercepted infrared thermal image of the single insulator, two image segmentation threshold extracting methods are presented. One extracts the segmentation threshold from the histogram envelope line, and the other gets the segmentation threshold by the method of OTSU in logarithmic transform domain, based on the two threshold extracting methods, a segmentation method integrated threshold segmentation and morphologic post-processing is presented for insulator infrared thermal image. Experiment results indicate that the segmentation quality is eminent, the insulators are intact and their margins are clear.4. Because of the insulator infrared thermal images interlapping with each other in the insulator strings, half of the disc surface of insulator is the region of interest in the research. The validity of the feature extraction directly depends on whether the half of the disc surface could be well segmented from the image or not. The disc surface image of insulator is characteristic of ellipse. The edge points of the disc surface of insulator are sampled through different angle’s straight line extending from the barycentric coordinates which are computed from the segmented image. The elliptic equation of the disc surface edge is fitted by the least square method. The ellipse image region above its long axis is abstracted, which is the half of the disc surface of insulator. Experiment results show that the presented method can obtain the half of the disc surface of insulator uniformly and normatively.5. The difference of the real temperature and the measured temperature is unpredictable by reason of the temperature measurement error of the infrared imaging system. To avoid the error effect of the measured temperature and utilize adequately the measurement precision of the infrared imaging system, a pollution feature extraction method based on relative temperature is put forward to bring the temperature distribution more reliable and accurate. Pollution features are extracted on the basis of the heating distribution on the insulator surface. Pollution level recognition takes the influence of environmental humidity into consideration. Six statistical parameters, namely, the average, the variance, the skewness, the kurtosis, the energy and the entropy of the relative temperature distribution, are extracted as pollution features from the point of view of the difference of whole temperature distribution, and a back-propagation neural network classifier is designed to check the insulator pollution level. The radial mean values of relative temperature are extracted as pollution features for the difference of temperature distribution along the disc diameter, and insulator pollution level is evaluated by minimum distance classifier under the nearest humidity condition. The gray histogram of insulator infrared thermal image indirectly embodying the relative temperature distribution, the normalized gray histogram is extracted as pollution features, and site pollution severity class is evaluated by maximum comparability criteria of grey synthetically relational degrees under the nearest humidity condition. Experiment results prove the feasibility and effectiveness of the three proposed methods.6. The best visual angle is propitious to improve the accuracy of detecting insulator pollution level by infrared imaging. A method to determine the best visual angle is proposed through comparative analysis of the same pollution features abstracted from insulator infrared images with Fisher criterion. Experiment results indicate that the thermal field of insulator surface significantly changes with the angle of view, and the features of lower surface have better classification performance to the uppers. It is recommended that the visual angle should aim at the lower surface for insulator pollution level detection by infrared thermal imaging.7. To acquire pollution features with excellent classification performance and lower characteristic dimension, a pollution feature selection method based on single factor variance analysis is brought forward. Experiment results show that the proposed method is simple and effective, not only decreases the complexity of data processing, but also avoids the undesirable characteristics into the feature subset for classification, improves the accuracy of pollution level classification.To sum up, key techniques of pollution level detection of insulator strings using infrared thermal imaging have been resolved in this paper. It is able to realize an accurate detection of pollution level of insulator strings by infrared thermal imaging.更多还原