Abstract
Pollution flashover and discharge caused by pollution deposition on the surface of insulators often occur in the field of power system. In the natural environment, insulators are affected by different weather and pollution sources, and the pollution deposition on the surface dissolved in water damages the insulation performance of insulators, and then leads to power failure of multiple transmission lines when pollution flashover accidents are serious, with serious consequences. Research and investigation have proved that the types of pollution components on the surface of insulators have great differences in their contribution to the pollution flashover phenomenon. In order to carry out regular cleaning of insulators, the identification and detection of pollution components on the surface of insulators is of great significance to predict insulator flashover voltage and prevent pollution flashover accidents. In this paper, based on hyperspectral image technology in the laboratory under ideal conditions to carry out the integrated pollution insulation uncleanness identification analysis experiment, established the insulator samples of hyperspectral image database, set the insulator impurity composition identification model based on deep learning, validate the convolutional neural network model for insulation slices of impurity composition to identify the feasibility and effectiveness.