Research On Substation Equipment Recognition And Application Based On Deep Learning

With the rapid development of my country’s power industry,substations,as the most core node in the transmission of power resources,their safety and stability are of vital importance.The prerequisite for ensuring their safe and stable operation is to conduct regular inspections of substation equipment.With the enhancement of computer hardware performance and the rapid development and maturity of machine learning,deep learning,target detection and other technologies,the inspection work of substation equipment is gradually becoming automated and intelligent.This paper mainly studies the application of deep learning-based target detection algorithms in substation equipment identification.The purpose is to realize intelligent detection and recognition of common equipment in substations-signal lights,digital display instruments,pointer instruments,etc.,so as to assist workers or even replace them.Manual inspection of equipment,the main work of this article is as follows:First,use denoising and enhancement algorithms to improve the image quality of the equipment images collected by the substation,and reduce the impact of image noise on subsequent equipment identification;design the substation equipment identification method,using the method of combining the largest connected area and the smallest external rectangle to locate the instrument Pointer position and get its angle information.Secondly,in view of the shortcomings of traditional algorithms in meter recognition,this article uses Faster-RCNN to classify and locate substation equipment,and then combines traditional target detection algorithms to achieve the reading of logarithmic display meters and pointer meters;for meters The characteristic of pointer rotation is indeterminate.On the basis of Faster-RCNN,it is proposed to use RRPN to detect the rotation of the pointer of the instrument,so that the bounding box can better enclose the pointer;RRPN has the problem of slow detection speed and low recall rate.Single-stage deep learning is studied.The algorithm RetinaNet detects and recognizes substation equipment,and proposes to use the rotating RetinaNet detection algorithm to recognize the pointer of the instrument,which greatly improves the detection accuracy,recall and speed.According to the relatively large width and height of the meter pointer,Pio U Loss is used to calculate the frame regression loss value based on the rotating RetinaNet algorithm,which further improves the positioning accuracy of the meter pointer.Finally,in order to solve the problem of too few data sets,the use of Generative Adversarial Networks combined with traditional rotation,mirroring,flipping and other methods to increase the number of equipment images,effectively expanded the data set;designed and developed a substation equipment identification system,Intelligent monitoring and early warning of substation equipment.