The Research Of Intelligent Diagnosis For Corona Discharge Defects In High Voltage Lines Based On Bi-spectral Image Fusion

Corona discharge is one of the early phenomena of defect fault,which usually occurs in the defective position of insulator,grading ring and conductor of high voltage power.If not detected in time,these defects not only cause a lot of power loss,but also lead to electromagnetic interference,accelerate the aging of power equipment,lead to power failure accidents,fire and other hazards.Therefore,it is very important to detect the corona discharge phenomenon.Bi-spectral ultraviolet(UV)detection is one of the most advanced corona discharge detection technologies.It has high sensitivity and can accurately and reliably detect weak corona discharge in power system.However,at the present stage,the Bi-spectral UV corona discharge images rely on the detection personnel to diagnose the defects on the spot.There are some problems in this kind of operation,such as inaccurate evaluation of corona discharge intensity,lack of defect reference standards and objectivity,low diagnosis efficiency and fatigue of personnel.On the basis of CDH02 Bi-spectral UV corona detector developed by our research group,the intelligent diagnosis algorithm of corona discharge defects is studied in this paper.Through the investigation of UV imaging detection,visible image recognition and information fusion,an intelligent diagnosis algorithm for corona discharge defects is proposed,which analyzes the intensity of corona discharge with UV image features,identifies the types of corona discharge parts with convolution neural network technology,and fuses multi information to diagnosis based on DS evidence theory.This thesis focuses on the following contents.The analysis method for corona discharge intensity based on equivalent photon number is studied.At present,the photon count is used as the discharge intensity standard in the current UV detector,but the UV photon count lacks the traceability standard,and the image characteristics are weak related to ultraviolet radiation.Aiming at this problem,the radiation standard traceable to NIST in the United States is established,and the calibration of the UV energy of ICCD is completed.Based on this,this thesis proposes a corona discharge analysis method,which calculating the number of equivalent photon according to the sum of the gray of UV spots area.In the future,a clear risk level standard for corona discharge defects can be formed based on the inspection experience.The identification method of power component based on convolutional neural networks(CNN)is studied.Firstly,a complete image data set of insulator,grading ring and conductor is established,which provides the data benchmark for corona discharge parts identification.On this basis,the Residual Networks(Res Net)recognition models with different convolution depths are trained.Then a simplified residual structure is proposed to shorten the training time of the Res Net.Finally,a fractional downsampling method with more flexible output layer size is proposed.Based on this,a fractional down sampling Res Net with more intermediate levels is designed,which improves the accuracy of grading rings.Fractional down sampling Res Net can process 987 power component image blocks per second,and the average accuracy of power component recognition is up to 98.9%.It meets the requirements of accuracy and real-time of power component identification,and greatly improves the efficiency of corona discharge defect diagnosis.The multi information fusion diagnosis method based on D-S evidence theory is studied.The task of corona discharge defect diagnosis needs to comprehensively judge the severity of corona discharge defects according to component type,corona discharge intensity and historical inspection experience.Based on the experience of corona discharge inspection and the method of equivalent photon number,the basic probability assignment(BPA)models of corona discharge defects of insulator,grading ring and conductor are established by using D-S evidence theory.Then,an intelligent diagnosis algorithm for corona discharge defects is proposed based on weighted fusion optimization method.The algorithm is simulated and analyzed by the patrol image data at the end.The results show that the intelligent diagnosis algorithm can clearly represent the possibility of defects.Compared with manual diagnosis,this method is faster and more reliable.The algorithm improves the efficiency and reliability of corona discharge defect diagnosis.Through the above three aspects of research,this thesis established an accurate and stable analysis method of corona discharge intensity,realized the accurate and high-speed identification of discharge parts,and proposed an objective and efficient intelligent diagnosis algorithm of corona discharge defects.The research results provide a certain help for the realization of high voltage power corona discharge defect intelligent diagnosis.