Research On Grid Insulator Identification Technology Based On Machine Learning

With the rapid development of intelligent power systems,computer vision technology has been widely used in online monitoring and inspection of power equipment.Insulators are important transmission line components in the entire power system and power grid.Due to its long-term exposure in the wild,the failures caused by foreign objects such as hanging,cracking,breakage and self-explosion frequently cause serious impact on the reliability,safety and normal operation of the power transmission line.Therefore,the positioning,condition monitoring and fault detection and identification of the grid insulators can help improve the effectiveness,safety and automation level of the transmission line inspection.This thesis combines current research status and technical progress at home and abroad,based on the image recognition theory and the key technology of grid insulator identification,insulator image data set for machine learning is acquired by a drone equipped with a camera,and pre-processing the collected grid insulator image to extract the insulator characteristics of the machine learning.In the process of grid insulator identification,this thesis effectively combines YOLOv3 and insulator 3D model and skeleton extraction with AdaBoost,and proposes two sets of identification methods,namely AdaBoost algorithm based on YOLOv3 and AdaBoost algorithm based on 3D model and skeleton extraction.The whole identification process of grid insulators is analyzed,and the results of two identification experiments based on machine learning for grid insulators under different identification principles,methods and the same identification target are obtained.Among them,the AdaBoost algorithm based on YOLOv3 is to locate the insulators in the image by target candidate region selection,target candidate region effective aggregation,and classifier recognition step-by-step.The AdaBoost algorithm based on 3D model and skeleton extraction is based on grid insulator 3D modeling,skeleton extraction and classifier identification to achieve accurate and effective positioning of grid insulators.The experimental results show that two sets of solutions effectively improve the effect of identifying insulators.Compared with the YOLOv3-based AdaBoost algorithm for detecting the location and recognition experiment,the 3D model and the skeleton extracted AdaBoost algorithm are better for grid insulator identification experiments,the identification method has better recognition effect,and the detection rate measurement parameter F1 is up to 95%,which has high practical value,and can provide basis for automatic inspection,fault diagnosis and identification processing of power system and smart grid.