基于多光源图像信息融合理论的绝缘子污闪预测研究

The pollution flash over on the insulators makes serious threat to the safe of high voltage transmission line. Therefore, it is an important content of the smart grid construction to explore the non-contact method of recognizing the status of insulator pollution to realize the accurate, efficient, and safety inspection of the pollution status. In this project, on the one hand, by using vision monitoring camera, thermal infrared imager and ultraviolet imager, we gets multi-source image of the operating insulator which can reflect the status of polluted insulator from various aspects, such as the color difference, temperature and intensity of discharge. Suitable fusion strategy for multiple source images , the three types of image features in different light sources complement each other. After fusing the information in feature level and decision level respectively, we can realize the accurate recognition of insulator pollution status in multiple light sources images. On the other hand, the theory of the development process of insulator pollution flashover is studied, forming the characteristics which can accurately describe the various stages such as insulator surface pollution, wet, dry area formation, partial arc formation and so on, we can establish the insulator pollution status database which is capable of self-studying and accumulating knowledge. On the basis of the study before, we can study the association between the pollution flash over mechanism and multi-source image characteristics, and use the visible light, infrared and ultraviolet image characteristics to describe the process of pollution flash over. Then setting up the quantitative relationship between the parameters in different stages of the pollution flash over and the multi-source image characteristics, using multi-source image characteristics to describe the development process of pollution flash over, we can achieve insulator pollution status intelligent evaluation based on multi-source image information fusion at last.

绝缘子污秽闪络对高压输电线路安全构成严重威胁，探索绝缘子污秽状态有效识别方法，实现绝缘子污闪预测，是智能电网建设的重要内容。本项目一方面借助可见光相机、红外热像仪、紫外成像仪获得绝缘子的多光源图像，从色差、温升和放电强度等不同角度反映绝缘子污秽状态，探索适用于多光源图像的融合策略，以不同光源的三种图像特征互补，分别在特征级与决策级进行信息融合，实现多光源图像绝缘子污秽状态的准确识别。另一方面，进行绝缘子污闪发展过程的理论研究，形成可准确描述绝缘子表面积污、潮湿、干区形成、局部电弧产生等各阶段机理的特征量，建立具有自学和知识积累能力的绝缘子污秽状态数据库。在此基础上，进行污闪机理与多源图像特征的关联研究，利用可见光、红外、紫外图像特征来表征污闪发展过程，建立污闪各阶段参数与多源图像特征间的定量关系，用多源图像特征描述污闪发展过程，在多源图像信息融合基础上实现绝缘子污秽状态智能评估。

绝缘子污秽闪络对高压输电线路安全构成严重威胁，探索绝缘子污秽状态有效识别方法，实现绝缘子污闪预测，是智能电网建设的重要内容。但目前污秽绝缘子直接检测方法难以推广应用，而间接检测方法可靠性不高。本项目通过绝缘子污闪机理分析与多源图像检测技术结合，综合利用可见光、红外及紫外图像信息，从污秽层颜色、温升及放电强度等角度对污秽状态进行较全面的描述：（1）对绝缘子污闪机理进行了研究，提出了绝缘子表面临界击穿数学模型与绝缘子污闪概率估计数学模型，计算和验证了绝缘子表面起始放电电压与不同条件下的绝缘子污闪概率；（2）进行了绝缘子图像特征与污秽度的关联研究，建立了绝缘子表面光路模型、发热模型和光子辐射模型，分析了绝缘子表面颜色分量、干燥带温升、最大辐射光子数等特征参数随绝缘子污秽度的变化规律；（3）获取了绝缘子可见光、红外及紫外图像，采用图像处理、特征提取、特征选择等技术，得到了能够有效表征绝缘子污秽状态的三种图像特征量，并分析了各图像特征量随环境因素的变化规律；（4）提出了基于变权值的多光源图像决策级信息融合方法，依据三种图像检测方法的适用范围，实现了各湿度环境下的绝缘子污秽状态准确识别，并以绝缘子污闪概率为依据进行了绝缘子污闪预测与示警；（5）提出了基于轮廓信息的异源图像的配准方法和多光谱图像重构方法，重构成了同时反映绝缘子污秽状态的色彩、温升、放电强度等三方面特征的图像，实现了绝缘子污秽状态检测的可视化；（6）进行了绝缘子污闪机理与重构图像特征关联研究，分析并提取了绝缘子污闪发展各阶段的重构图像特征，实现了基于重构图像的绝缘子污闪发展阶段识别和污秽状态识别。本项目研究成果提升了图像检测准确率和检测效率，推动了图像识别方法在绝缘设备缺陷检测技术上的应用。.综上，本项目已完成计划书所有预定研究内容，发表SCI/EI论文18篇，申请发明专利11项（其中授权2项），获得软件著作权1项，研究成果超过预期。