Research On Non-contact Detection Method Of Insulator Pollution Degree And Components Of Transmission-line Insulators

The contamination degree and composition of insulator are the direct correlation parameters which affect the pollution flashover characteristics.Therefore,the accurate and rapid detection of contamination is of great significance to prevent pollution flashover.At present,the traditional measurement methods adopted by power department need offline detection,so it is difficult to meet the development needs of intelligent and data-based power grid.Therefore,based on the characteristics of natural pollution spectrum and engineering problems,this paper proposes a non-contact rapid detection method.The traditional classification modeling approach has limited recognition capability and low generalization performance.Based on the analysis of the spectral characteristics of natural pollution,a convolutional neural network is introduced to build a binary classifier structure model to realize the identification of the main components of mixed pollution.The results show that the model can identify the typical pollution components（Al（0H）₃,Si O₂,Na Cl,Ca SO₄）with an accuracy of 91.6%.The existing quantitative pollution detection methods ignore the correlation characteristics between soluble and insoluble materials,resulting in the low applicability and accuracy.Based on the idea of interval prediction,a deep extreme learning machine algorithm is used to realize the quantitative detection of transmission line insulator pollution degree in layers.The comparison results show that the proposed method improves the prediction coefficient R2 from 0.5457 to 0.9146,and reaches 0.9544 after combining feature screening and parameter optimization;the accuracy of field test is 88.89%,which verifies the good engineering practical value.Focusing on the problem of low sensitivity of pollution components detection caused by the interference of substrates on the spectral information,a method for removing substrate information from transmission line insulators based on convolutional noise reduction auto-encoders is proposed to achieve spectral correction.The pollution verification results show that the proposed correction method can effectively improve the spectral information interference of the composition,and the accuracy rate is increased from 33.33%to 83.33%.The above method can lay a certain theoretical foundation for on-line insulator detection on site.