Application Of Insulator Contamination Fault Diagnosis Technology Based On Optimized Random Forest

This paper mainly studies the image preprocessing of transmission line insulator,the identification and location of various insulators,the discrimination of insulator pollution degree and the classification of pollution level.Firstly,the acquired image is preprocessed to eliminate the interference of external light,fog and unclear image.Then,the image is identified and located by using yolov4 deep learning algorithm.Then,the image is segmented by grabcut to get the part of the insulator image.Then,the texture feature,color feature and SVD feature are extracted from the image,In order to extract useful image information is the premise of insulator pollution detection.Then,according to the extracted features,the pollution discrimination function is realized.Insulator pollution detection technology is the process of data mining and processing.The insulator image features are predicted and classified,and the final insulator pollution level is obtained on the basis of optimizing random forest algorithm.1)In this paper,the image of uneven illumination,unclear,fog and other external environment is preprocessed;The algorithm is improved.The improved backlight algorithm based on fusion is used to deal with the uneven illumination environment,and the defogging algorithm based on sky detection is used to deal with the foggy environment.2)In traditional random forest algorithm,for small data or low dimensional data(data with less features),it may not produce good classification and slow calculation speed.The algorithm calculates the optimal tree depth when the number of trees is set to regenerate the random forest,and then calculates the confusion matrix accuracy(CMA)of each random decision tree generated.By comparing the CMA value,the decision tree with the highest value is selected for similarity comparison,and the decision tree with the highest final index and the lowest similarity is selected as the random forest model,The final accuracy is obtained.Under the classification and prediction of the algorithm,the detection accuracy is improved,and the correctness and feasibility of the algorithm are verified by theory and experiment.3)According to the national standard GB / T 4585-2004 and DL / T 859-2015 artificial pollution experiment,more than 300 insulator images with different pollution levels are obtained by solid layer method,and the insulator pollution image is segmented by grabcut,and then the texture,SVD features and color features of the target segmentation image are extracted,and the feature dimension is reduced by KPCA algorithm,It provides the basis for the classification and prediction of the subsequent optimization random forest algorithm.Finally,the useful database is applied to the real environment to verify the feasibility of the algorithm,and provide a theoretical basis for other research in this area.In this paper,binocular camera and other equipment and kaolin,sodium chloride and other experimental components were used in the hardware and experimental facilities.In the implementation of the algorithm,the software is used to verify the improved algorithm to ensure the feasibility.The experimental results show that,on the basis of the algorithm proposed in this paper,the accurate recognition and preprocessing of insulator image are better,and the final accuracy rate of insulator pollution level discrimination is increased by2%-4%,which proves the feasibility of the method proposed in this paper,and makes a modest contribution to improve the efficiency of insulator pollution level detection technology.