Design And Implementation Of Switch State Indentification System For Substation Based On Deep Learning

With the expansion of substation scale in China,it is more and more difficult to monitor all kinds of equipment in substation.Due to the frequent use and maintenance of switching equipment in substation,it is very difficult to test substation switch equipment only by manual maintenance.Therefore,taking the secondary switching equipment of substation as the research object,this paper proposes a method about identifying the state of switching equipment based on deep learning,which can automatically detect and identify the state of switching equipment,and has important research significance and application value for realizing the intelligent substation system.Faster R-CNN algorithm is a classical algorithm applied in the field of the target recognition by deep learning.Region proposal network is proposed to generate high-quality and small number of region proposal boxes,which not only improves the detection rate of the model,but also increases the recognition accuracy of the model.This paper proposed four improvement strategies for Faster R-CNN algorithm: the data amplification method,the feature fusion method,the region proposal boxes clustering method and Soft-NMS post-processing method.Among them,the data amplification method mainly solves the problem that there are few sample data sets that can be collected at the present stage.The feature fusion method is to combine feature information such as shallow contour and color,with deeper and more abstract information to form new features,which aims at the characteristics of small and difficult to identify switch targets.The region proposal boxes clustering method is to analyze the size of the region proposal boxes based on the regularity of the shape of the switch target,so that the size of the region proposal box is more consistent with the actual size of the switch target.The Soft-NMS post-processing method replaces the original NMS method,and effectively solves the problem of suppression of adjacent regions with high overlap rate and reduces the number of false positive samples.Through the design and implementation of the stripping experiment of the improved strategy,the feasibility and effectiveness of the four improved strategies in the substation switching state identification are verified.At the same time,a comparison experiment was conducted between the improved strategy and the existing Fast R-CNN algorithm and Faster R-CNN algorithm.The experimental results showed that the recognition accuracy of the Faster R-CNN algorithm improved strategy reached about 93%,which had significant advantages in substation switching state recognition.Finally,based on the trained recognition model,substation switch state identification system was designed and implemented,mainly by the user management,image upload,image recognition and identification information management and data sets of five modules,indirect realized the recognition model of visualization,and provides remote monitoring for substation workers substation switch equipment services,to further advance the process of intelligent substation.