Research On The Key Technology Of Transformer Fault Diagnosis Based On Electrical-graphic Information Fusion

With the development of infrared thermal imaging technology,more and more substations use this technology to detect faults in transformers.Due to the low imaging quality of infrared images and the existence of many interference factors in the images,it is difficult to obtain good results in transformer fault detection using only image processing.Using the operation data to monitor the transformer operation status,but only using this technology can’t further identify the fault type and extract the fault point.Combining the two methods can diagnose the fault of the transformer more accurately and comprehensively,and improve the construction level of the smart grid.This paper studies a fault diagnosis method combining the operating data of the transformer and the infrared image.In fault detection,by studying the characteristics of transformer operating data and infrared images,this paper conducts research through three aspects: judging whether the operating data is abnormal,identifying fault types and extracting abnormal hot spots.1)To analyze the characteristics of the transformer operation data,in order to maximize the difference between the features,the original 29 features were reduced to two features using principal component analysis(PCA).The support vector machine(SVM)algorithm is used to analyze and detect the transformer operation data,and the real-time operation status of the transformer is obtained.2)Real-time infrared images are acquired for the time points where faults occur and fault analysis is performed on them.Since the infrared image dataset is small,this paper will use migration learning to increase the training depth,extract the transformer fault features by Res Net algorithm,load the model weight file pre-trained by Res Net on large dataset,and add the infrared image data for training to get the fault type recognition model.3)The transformer infrared image is segmented using the improved watershed algorithm to avoid interfering factors in the background outside the device,and the location of the hot spot is obtained by converting the color characteristics exhibited by the transformer infrared image to the HSV color space by converting the RGB color space to the HSV color space,extracting the abnormal hot spot whose color is close to white,and performing as the point with the highest brightness in the HSV color space.In this paper,a transformer fault diagnosis model with fault type identification,abnormal hotspot location and multi-information fusion is constructed based on the basic status information management system of power system operation through the feature analysis of transformer operation data and infrared images.This model can be used in equipment overhaul and maintenance work in the power system to realize analysis of transformer operation status and fault warning,which improves the accuracy of fault warning.It realizes non-contact and non-disconnection fault detection,which guarantees the personal safety of maintenance personnel and property damage.Through this fault diagnosis means will reduce the equipment in the use process of major failure accidents,improve the service life of the equipment.