Analysis And Evaluation Of Insulator Operation Status Based On Infrared And Ultraviolet Imaging Detection Technology

Under long-term electric,thermal,environmental and mechanical stresses,transmission and transformation insulators inevitably suffer from insulation degradation,aging,even forming defects,which threatens the safe and stable operation of power systems.The discharging and heating phenonmenon of insulators,to a certain extent,correlate with their operating status.Therefore,previous studies mainly utilized the ultraviolet imager and the infrared thermal imager to perform live detections on insulators.Relevant characteristic parameters were set or extracted,and the data analysis and the comparison with experimental phenomenon were carried out,in order to construct relevant diagnostic criteria.With the continuous development of artificial intelligence,the theory and methodology of deep learning and its typical applications is triggering breakthroughs in multiple industries.Therefore,based on the infrared and ultraviolet live detection technology for insulators,this thesis researches into the application of the image classification,the target recognition,the image segmentation and other artificial intelligence and deep learning algorithms.The results of this thesis aim to break the predicament of "the manual setting of complex algorithm flow" in the infrared ultraviolet detection,to realize the end-to-end model training and detection,and to promote the intelligent evaluation and diagnosis of the infrared and ultraviolet imaging detection.The main contents of the study are as follows:The infrared spectrum of deteriorated insulators were evaluated and diagnosed by three algorithms:the BP neural network,BOA-SVM and the convolution neural network.The thermal infrared detections of deteriorated insulators in transmission lines were carried out under multiple environmental factors.The thermal characteristics of deteriorated insulators were analyzed at various humidities,at different positions and with contamination.The infrared thermal imaging database of degraded insulators and normal insulators was constructed.The BP neural network algorithm was used to compare the training and detection effects of three characteristic parameters,i.e.the color histogram,the color moment and the center line color vector matrix,on degraded insulators.Bayesian optimized supporting vector machine classification and evaluation diagnosis algorithm was adopted to realize the categorized evaluation and diagnosis of the infrared images of degraded insulators.The convolution neural network,a machine learning model based on deep learning,was modified and trained to evaluate and to diagnose the infrared images of deteriorated insulators with increased accuracy.The algorithm of infrared thermal imaging insulator heating target recognition based on deep learning was studied.A large number of infrared images with abnormal heating were collected and were sorted out.All hot spots of abnormal heating in infrared images were labeled manually,and a trainable data set which conforms to the deep learning of the target recognition was constructed.Faster-RCNN and YOLO-V3 target recognition algorithms were used to realize the recognition and the frame selection of insulator abnormal heating target points,which can effectively shield the interference from non-fault hot spots that are negligible,thus shed light on a new path to realize the lean inspection of the infrared heating abnormality.The classification and evaluation of ultraviolet discharge spectra of insulators based on the convolution neural network was studied.The power frequency flashover test of ceramic insulators was carried out using FILIN ultraviolet discharge imager,and the spectra library of different discharge stages was established.The Alexnet deep convolution neural network model was optimized and was used to achieve intelligent classification and evaluation of ceramic insulators with high accuracy for ultraviolet imaging detection.This research provides a solution for the insulator flashover assessment and diagnosis based on the deep learning algorithm.The composite evaluation and diagnosis model of full convolution neural network and convolution neural network for ultraviolet contamination detection of insulators was studied.Using the ultraviolet images of the ceramic insulator contamination,the spectrum library of different status was established.The preprocessing algorithm of the full convolution neural network model for the images from the South Africa CoroCAM ultraviolet imager was studied to split and extract the main light spots and to filter out the background noises.Then,the diagnosis and evaluation of the ceramic insulators contamination by ultraviolet images was achieved via deep convolution neural network.Finally,a new method based on the FCN and CNN model was developed to evaluate the contamination of insulator strings.An integrated prototype of multi-path and multi-sensor was developed,which combines the integrated ultraviolet,infrared and visible cameras,consists of various micro-meteorological environment sensors and provides a more comprehensive solution for the correction of multi-path detection.The infrared and ultraviolet backstage diagnostic software for insulators was developed,which could be connected with the multi-path prototype equipment,endowing the multi-path detection equipment with preliminary diagnostic functions.The relevant diagnosis algorithms were deployed in the software,and the application of the infrared and ultraviolet imaging to the diagnosis and evaluation of insulators operation status was achieved.