Study On Classification Method Of Grounding Grid Corrosion Degree Based On Convolution Neural Network

The safe running of substation is an important guarantee for the normal operation of power system,substation grounding grid is the key to ensure the electrical equiprment,power system and personal safety,the grounding conductor of our country is mostly made of ordinary carbon steel,which is buried in the underground soil for a long time,and will gradually corrode,even break or be completely corroded,this is a serious threat to the safe operation of electric power system and the personal safety of the staff.Therefore,it is necessary to study the detection method of grounding grid corrosion.In this paper,the corrosion image of substation grounding grid is studied as the research object,the classification algorithm based on convolutional neural network,using the different characteristics of the corrosion appearance in the different levels of the standard GB/T6461-2002 as the basis of classification,to classify the degree of corrosion of the grounding grid.The main work is as follows:Firstly,the corrosion of the grounding grid is simulated,and the corrosion image database is built by using the collected images.In order to test the recognition ability of the CNN model in the complex scene,this paper establishes two kinds of contrast samples,which are the sample database and the original image database.The method of histogram equalization is used to enhance the contrast of the image,and the vector median filter is used to process the image.Secondly,the classification model of grounding grid corrosion degree is established.By improving the structure of the LeNet-5 CNN classic,is constructed in the study of CNN classification model,in order to find the optimal classifier for corrosion image classification,this paper set up Softmax and SVM two kinds of classifiers,and two kinds of CNN classification model in the pretreatment of samples and the original image sample library by contrast experiment the experimental results show that the classification performance of SVM is better than Softmax,CNN model in the pre-processing of the sample recognition rate is higher than the recognition rate of the original image,the traditional CNN model has a poor generalization ability for corroded images in complex scenes.Again,propose an improved model for the recognition of CNN model in the original image samples on the corrosion rate is low,the improved model will be classified image input at the same time the two CNN model for feature extraction,and two kinds of feature extraction model merging into the classifier.The experimental results show that the recognition rate of the improved CNN model is obviously higher than that of the traditional CNN model,and the classification effect of the improved CNN model with SVM is the best.Finally,the Lab VIEW and MATLAB are combined to develop the grounding grid corrosion degree classification system.Data storage,model building,model testing and other functions are realized.