Image Analysis Of Glass Insulator Icing

Transmission lines are facing severe challenge both physically and electrically under icing conditions,which poses a threat to safe operation of power systems.It is an important subject of online monitoring for transmission lines that how to accurately and effectively monitor icing conditions on transmission lines to prepare for timely preventive and ice-meltingmeasures.In recent years,with progress in image processing and analysis,as well as the development of machine learning,power grids have become more intelligent.Since the severe ice disaster of China southern areas in 2008,China South Power Grid quickly established Transmission Line(Icing)Online Monitoring System to monitor icing conditions of transmission lines,which collected a large number of images on transmission line icing,as well as meteorological and mechanical data.The thesis conducted image analysis on of icing glass insulators.Lots of glass insulator icing images were researched on and according to the observed ice forms and positions,icing on insulators can be categorized as Snow,Glaze,Heavy Rime,Rime,Light Rime,and Partial Rime.Glass insulators covered with the above six types of ice and these free of ice were analyzed through their texture features.This thesis also conducted research on glass insulator segmentation methods and compared the segmentation results of Threshold,Sobel,Laplacian,Canny,Seed Region Growing,and GrabCut method under different backgrounds and icing conditions.Following that,the thesis explored representation method of icing severity.Based on GrabCut method,Change Percentage of Image Area was calculated for icing glass insulator images and compared to artificially observed icing thickness.The thesis also carried out analysis on graphical changes of insulator shed gaps and overhangs before and after icing.To find the relationship between glass insulator graphical shed gaps and overhangs and icing severity,the thesis developed experiments to calculate graphical shed gaps and overhangs of insulators before and after icing by recognizing contour convex defect.The thesis provided a thought to utilize the unstructured and structured online-monitored data at the same time to realize prediction for states of transmission lines.First,it represented the unstructured data with structured forms and combine them with other structured data to get the original features,then,it applied Dropout neural networks to optimize the feature representations and obtain the patterns between the features and outcomes.