247 / 2018-09-24 19:54:06

Detection Method of Insulator Contamination Degree Based on Hyperspectral Technique

hyperspectral technique,insulator,contamination degree,multiplicative scatter correction,back-propagation neural network

The flashover of transmission line insulators is always a key problem to be solved in the safe and stable operation of power system. Therefore, the on-line detection of insulator contamination degree of transmission line is of great importance to the prevention and control of flashover. Traditional contact detection methods have certain limitations in operation and application, thus hardly realizing on-line detection of insulator contamination degree, while hyperspectral technique is a new kind of technology which combines image with data based on imaging spectroscopy technology, with some characteristics as multiband (up to hundreds of wavelengths), high resolution and image combining with spectral curves, etc. Its advantages lie in the rich information collected by hyperspectral images, high recognition and large amounts of data description models. This paper proposed a non-contact detection method of insulator contamination degree based on hyperspectral technique, which can realize the on-line detection of the contamination degree. Firstly, artificial contamination samples with different contamination degree were prepared by using the artificial contamination test method recommended by standard GB/T 22707-2008. Two groups of samples were prepared by using silicon rubber insulation sheet (with size of 5 cm*5 cm) as the substrate, and NaCl mixed with kaolin and CaSO4 mixed with kaolin were as the contamination components to cover the substrate. Secondly, hyperspectral testing platform were built to obtain the hyperspectral images of the samples with different contamination degree. The test was carried out under the condition with temperature (18-25℃) and humidity (30%-80%). Thirdly, in order to overcome the influence of image noise and dark current in the band where light intensity distribution is weak, the black-and-white correction and multiplicative scatter correction were applied to correct original hyperspectral images to enhance the relationship of spectral absorption information with components or contamination degree and to reduce the dispersion of spectral curves. After that, hyperspectral curves from the region of interest (ROI) of corrected images were obtained, and the sample image data was divided into training sample and test sample according to the statistical rule. Finally, contamination degree was divided into four levels, which were defined from “very light” to “heavy”. Multi-classification model of back-propagation neural network was built based on training sample data, which was designed to realize the contamination degree classification of test samples. The results show that contamination of different components appears different reflection characteristics, which varies differently with contamination degree; and the absorption peak, the position of reflection peak, amplitude and the change trend of the hyperspectral curve obviously change with different kinds of contamination on the surface of silicone rubber, while only the amplitude obviously changes with the same kind of contamination on the surface of silicone rubber; and the multi-classification model can accurately and rapidly classify the contamination degree, with the contamination degree classification accuracy of NaCl and CaSO4 reaching more than 90%; Therefore, this method can be applied to the on-line detection of insulator contamination degree, which can realize the rapid, accurate and undamaged detection, providing a new idea for the detection of insulator contamination degree.