Research On Anomaly Detection Technology Of Insulator In Traction Substation Based On Adversarial Generative Network

Insulator in traction substation is an important high voltage equipment in traction power supply system of high-speed railway.Its operation state will affect the safety and stability of traction substation.Therefore,it is of great significance and engineering value to research the anomaly detection technology of insulator in traction substation and realize the dynamic monitoring of insulators.This paper studies the traction substation insulator based on the generated adversarial network anomaly detection methods,firstly,the video image of traction substation is extracted,and the insulator in traction substation is annotated by image annotation tool.These original images and the image data generated by annotation constitute the insulator object location data set of this paper;Then,aiming at the problem of tilted insulators in reality,a rotating-frame detection model based on R~3Det network is constructed.The network model can effectively detect the tilted insulators,and then achieve the object location of the insulators.Then,the insulator image is extracted according to the insulator object location results.Aiming at the problem of insufficient abnormal insulator data in the actual situation,the abnormal insulator anomaly detection data set in this paper is constructed by manually making abnormal insulator image.Then,a generation adversarial network model is constructed,which overcomes the disadvantage of unbalanced distribution of positive and negative samples in the data set,and realizes the reconstruction of abnormal insulator data into normal insulator data,the anomaly score is designed to judge whether the input insulator image is abnormal.Finally,the difference features of the input insulator image and the reconstructed insulator image are extracted by the image structure similarity algorithm.The region with the largest contour area in the difference image is found by setting the threshold,and the abnormal region is located according to its coordinates to realize the anomaly detection of the insulator.The validity and accuracy of the proposed method are verified by experiments in this paper.Among them,the object detection experiment results show that the R~3Det rotating object detection method designed in this paper can locate the insulator in the tilted state with high accuracy.The results of anomaly detection experiments show that the anomaly detection method based on generative adversarial network designed in this paper can effectively detect several abnormal states of insulators and locate abnormal regions,the anomaly detection accuracy is 92.5%.