Research On Transformer Casing Monitoring System Based On Computer Vision

Transformer casing is an important annex of the transformer,its stability and reliability related to the safe operation of the entire power system.With the increase of the power demand and the aggravation of the environmental pollution,the accident of electric power systemcaused by the fault of the transformer bushing is more serious than ever.The on-line monitoring of transformer casing and real-time control of its working condition can effectively avoid the occurrence of safety accidents.This research has important theoretical significance and practical value.According to the current research situation,this paper presents a method of intelligent monitoring system of transformer casing based on computer vision.Using digital image processing technology,computer vision technology and intelligent monitoring technology,the degree of hydrophobicity and fracture of casing are identified.This paper describes the system structure and workflow from two aspects of hardware and software.Construction of development platform based on ASP.NET,MATLAB and Microsoft SQL Server software.The real-time image of the casing is obtained by the remote camera,and the image is transmitted to the server for image analysis,processing,judgment and recognition by the network.In addition,this paper makes use of the natural raindrop to identify the water repellent level,and brings the crack condition into the monitoring range.Compared with the traditional monitoring methods,the monitoring accuracy and the monitoring function is improved.The main research contents include:1.The image preprocessing algorithm based on localized adaptive histogram equalization ofhomomorphicFor transformer casing image acquisition,affected by noise,weather,illumination and other factors,resulting in monitoring accuracy is not high.We propose a local adaptive histogram equalization method for image preprocessing based on homomorphic resolution.This method is used to remove the noise and enhance the original image before the feature extraction,so that the visual effect of the image is improved obviously,and the detail information of the edge is highlighted,which creates the conditions for the subsequent accurate identification.2.The algorithm of casing disk and water drop image segmentation based on improved Canny operator detection image edge and mathematical morphology operation correctionIn the case of raindrop edge detection for hydrophobic monitoring,the traditional Canny edge extraction algorithm can not achieve the ideal segmentation effect because the threshold can not be selected independently.The algorithm of threshold selection and Canny operator edge extraction based on Otsu algorithm is proposed.The edge extraction process can adaptively select the image segmentation threshold.Aiming at the problem of partial watermarking caused by environmental factors in the extracted edge,the modified Canny operator edge image was repaired using mathematical morphology.The problem of discontinuity and pseudo edge of water mark is reduced.3.The algorithm based on improved SIFT feature image matching and maximum entropy threshold for casing and background image segmentationIn view of the poor adaptability and difficulty in image matching,the SIFT feature image matching algorithm is selected,which has strong adaptability to illumination intensity,translation deviation,scale,rotation and shooting angle.In order to improve the matching efficiency and accuracy,an improved SIFT feature matching algorithm is proposed.The feature vector is replaced by a circle instead of a square neighborhood region to reduce the dimension of the data.Using covariance and inverse cosine function instead of Euclidean distance to reduce computation.Improving the matching accuracy by using bidirectional matching and RANSAC algorithm.At the same time,in order to eliminate the influence of the background on the target monitoring,the maximum entropy threshold method is used to segment the transformer casing area successfully.4.The least squares support vector machine based on Particle Swarm Optimization for water repellent level recognition modelIn this paper,the evaluation criteria and the extracted characteristic quantities of each hydrophobic grade casing are given.The least squares support vector model and training sample library are established.And the particle swarm optimization algorithm is used to optimize the parameters.The steps of recognition are proposed.Compared with other intelligent recognition algorithms,it is proved that the model has high recognition accuracy and the recognition rate is 96%.5.The fracture condition identification model based on improved BP neural networkThe characteristics of the image are extracted,and the BP neural network model and training sample database are established.Aiming at the inherent defects of neural network,the paper introduces the method of combining momentum and adaptive learning rate adjustment BP algorithm to construct the intelligent recognition model.With the support of the design of the working condition recognition device.Compared with other algorithms,it is proved that the model has high recognition accuracy and the recognition rate is 93.7%.Finally,from the point of view of practical engineering,this paper expounds the effect of the operation on the spot,and analyzes the actual data.The effectiveness and practicability of the system are proved.It provides an important theoretical reference and technical means for further improving the safe operation of the transformer.