Aging State Evaluation Of Insulating Paper Based On Image Feature Recognition

Transformers are the pivotal equipment of the power system.They not only play a key role in energy transmission and conversion,but are also an important part of the grid-connected process of various distributed energy sources.Once the transformer fails,it will have a huge impact on the safe operation of the power grid and cause serious economic losses.According to statistics,about 15% ～ 20% of transformer failures are related to transformer windings and most of the problems on the windings are related to the aging of insulating paper.Therefore,it is very important to accurately grasp the insulation state of the transformer for its safe and stable operation.As the main insulation of the transformer,the insulating paper will be aged due to the combined action of electricity,heat and light during operation.Accurate and effective evaluation of its aging state is very important to improve the operation reliability of transformer,which can provide a reliable basis for condition based maintenance and operation maintenance of transformer.Therefore,the paper starts with the image features of insulating paper by carrying out the aging experiment of insulating paper,and proposes image features to characterize the aging state of insulating paper,which solves the problem that the aging state of the main insulation of the transformer is difficult to directly detect on site and cannot can not realize rapid non-contact detection.The main innovative work carried out in this paper is as follows:(1)The accelerated thermal aging experiment of insulating paper was carried out,and the microscopic and molecular images of insulating paper at different aging stages were collected by optical microscope and scanning electron microscope,respectively.Through the analysis of different images,it was found that the long-term high temperature caused the fiber structure of the insulating paper to be destroyed,resulting in holes and gaps,which explained the complexity of the microscopic image texture with time.(2)Considering the influence of the laboratory environment on the collected images,the microscopic images are preprocessed,and the preprocessing effect is verified by the grayscale surface.Using box dimension and multi-fractal spectrum in fractal theory,three kinds of fractal texture eigenvalues related to insulating paper images are obtained.The fractal texture feature value is used as input,and the aging time of insulating paper is used as output,which is substituted into the extreme learning machine for classification and verification,and it is confirmed that the fractal texture feature value can be used to judge the aging time interval of insulating paper.(3)In order to study the relationship between the insulating paper image and the degree of polymerization of the insulating paper,the gray level co-occurrence matrix method and the gray level run-length matrix method in the statistical analysis were used to extract the texture feature values of the insulating paper image.The redundant feature values are excluded by the Pearson coefficient method,and the texture feature values with high correlation with the degree of polymerization are screened out.The screened texture feature values are used as input,and the aging time of insulating paper is used as output to be substituted into the support vector machine for classification and verification.After verifying the effect,the relationship between eigenvalues and aggregation degree is established by multiple linear regression method.