The discrimination method as applied to a deteriorated porcelain insulator used in transmission lines on the basis of a convolution neural network
Autor: | Zhang Kaiyuan, Fu Weiping, Yin Zihui, Pei Shaotong, Yunpeng Liu, Ji Xinxin |
---|---|
Rok vydání: | 2017 |
Předmět: |
Computer science
business.industry 020209 energy 020208 electrical & electronic engineering Feature extraction Insulator (electricity) Pattern recognition 02 engineering and technology Convolutional neural network Electric power transmission Fuselage Robustness (computer science) 0202 electrical engineering electronic engineering information engineering Infrared heater Artificial intelligence Electrical and Electronic Engineering business Gradient descent |
Zdroj: | IEEE Transactions on Dielectrics and Electrical Insulation. 24:3559-3566 |
ISSN: | 1558-4135 1070-9878 |
DOI: | 10.1109/tdei.2017.006840 |
Popis: | Based on the analysis of the principle and structure of a convolutional neural network (CNN) model used for in-depth learning, an intelligent discriminant diagnosis method for porcelain fuselage insulators in transmission lines is proposed. Firstly, the infrared image of a porcelain insulator is extracted, and then Lenet is used to optimize the network structure. Finally, the model of fixed parameters is formed by training. The model has high classification and judgment robustness and offers accuracy under different conditions such as: temperature, humidity, position of deterioration on the insulator, and thermal load, which allows weight-sharing in the CNN model under different environmental conditions. Based on the experimental data from an infrared heating experiment using a porcelain deteriorated insulator, this work uses the back-propagation gradient descent method to train the model, to form an intelligent detection model for deteriorated insulators. This method has the advantages of high accuracy and robustness, and represents a new method for intelligent detection of deteriorated insulators. |
Databáze: | OpenAIRE |
Externí odkaz: |