Research On Insulator Contamination Degree And Component Based On Hyperspectral Image Processing Technology

The insulators' surface on transmission lines will pile up a lot of contamination if being exposed to the air for a prolonged period,thus changing the external insulation strength,which may lead to contamination flashover.Therefore,it is critical to research the insulator contamination for the normal operation of transmission lines.Meanwhile,hyperspectral image possesses the properties of wide spectral range,image and spectrum,which can reflect the material microscopic characteristics.As such,hyperspectral imaging system is used to obtain hyperspectral image of the contaminated insulators in a non-contact way,and their contamination degree and component are further explored in this thesis.The specific research contents are described as follows.(1)Some research methods of insulator contamination based on hyperspectral image processing technology are proposed.Specifically,in order to realize the non-contact prediction of insulator contamination,this thesis first establishes the prediction model of insulator contamination based on support vector machine and partial least square regression.At the same time,in view of the learning efficiency and generalization performance of the model,this thesis built an insulator contamination prediction model based on weighted extreme learning machine.These models are applied to the hyperspectral image of contaminated insulators,and the satisfactory prediction performance is obtained,which verifies the feasibility of hyperspectral image processing technology for predicting the contamination degree of the insulators.(2)A new method for analyzing insulator contamination component is proposed based on the weighted kernel non-negative matrix decomposition(WKNMF).Because of the nonlinear mixing effect in the natural fouling process and the different influence caused by different pixels on the polluted component and mixing proportion of insulators,the WKNMF algorithm is employed to nonlinearly decompose the observation data,and each material component and its mixing proportion are obtained.Besides,the USGS(United States Geological Survey)database and the established spectral set of insulator contamination component are used to synthesize hyperspectral data based on the generalized bilinear model,respectively.The effectiveness of the nonlinear unmixing algorithm based on WKNMF is verified on the synthetic and real hyperspectral data.The experimental results show that the algorithm not only achieves the satisfactory performance,but also lays the foundation for analyzing insulator contamination component with hyperspectral nonlinear unmixing algorithm.