Research On The Anomaly Detection Method Of Insulator Contamination Level Based On Image Information And Improved Support Vector Data Description

The speed of filth falling on insulator’s surface is getting faster and faster because of the worsening air quality,directly or indirectly.So,the possibility of insulator flashover is also improving.So,detection and maintenance of external insulation properties of insulators is becoming more and more important.Some existing products and technologies in the market,especially the hot washing technology,have been widely applied and developed in the aspect of insulator’s external insulation capacity maintenance,because of their unique advantages.But,there is no perfect on-line detection method of insulator contamination,which provides necessary precondition and guarantee for the development of anti-fouling work.With the continuous development of science and technology,the demand for smart grid is more urgent,therefore,it is more necessary and urgent to study the online detection of insulator contamination level.Now,most of the on-line detection methods of insulator contamination level is based on the phenomenon of leakage current,luminescence,heating and voicing,which occurs when pollution flashover happens.However,most of the environment are bad when the pollution flashover accident occurs,so the traditional detection method is disturbed by many factors and the working environment is very poor.For the early warning technology of insulator pollution flashover," nip in the bud " is one of the best solutions,and not detected in the process of flashover.So finding a kind of detection method is a problem need to be solved,which can detect the pollution flashover before the accident happened.Visible light image is just such a new detection method and it does not need to use traditional electrical parameter or other expensive,heavy and precise equipment.In this paper,a kind of anomaly detection model of insulator contamination level based on improved support vector data description and image information has been established through image recognition and machine learning technology,the training samples were obtained by artificial pollution experiments.The main research contents of this thesis are as follows:1)First of all,feasibility and principle about the visible light image method applied to insulator contamination level detection is analyzed in detail in this paper,and it also points out the most basic requirements of contamination level detection model and some problems needing attention in the modeling process.2)Secondly,an improved support vector data description method is proposed in this paper in order to meet the basic requirements of insulator contamination level detectionmodel.And the advantages and disadvantages of the method are tested,through theoretical analysis,artificial data set experiment and standard data-set experiment.3)Then,the insulator image samples were obtained by artificial smear experiment,and the training data with low dimension is obtained by the method of image segmentation,color feature method,texture feature statistics,and feature fusion and descending dimension.4)Finally,the insulator contamination level anomaly detection model based on image information is established,by the improved support vector data domain description method,compared with support vector data description method,the disadvantages and advantages of this method are pointed out.And further,it describes the necessity and feasibility of online learning,briefly.