An Advanced Detection Method of Pollution Level of Transmission Line Insulators Based on GPLVM-IMVO-BPNN

Abstract

The ultra-high voltage direct current (UHVDC) transmission project plays a crucial role in global energy interconnection. As a key insulating component in transmission lines, silicone rubber insulators are prone to the deposition of pollution on their surfaces during long-term operation, which can lead to flashover and even cause large-scale power outages in grid. To achieve a high-precision assessment of pollution level on insulators, a non-contact detection method is proposed in this paper based on hyperspectral imaging (HSI) technology and back propagation neural network (BPNN). First, a hyperspectral imager is used to extract spectral data from the surface of the transmission line insulators. Then, four different dimensionality reduction methods are compared, and the Gaussian process late variable model (GPLVM), which provides the best spectral feature extraction performance, is applied to reduce the data dimensionality. Next, the BPNN, which is capable of characterizing complex nonlinear mapping relationships, is selected as the classification model. By incorporating an improved multi-verse optimization (IMVO), the optimal weights and biases for the model are obtained, enhancing the performance of global optimization and local exploration. The results show that the GPLVM-IMVO-BPNN classification model proposed in this study achieves an overall accuracy (OA) of 96.67%, which is a 7.92% improvement compared to the full band model. Additionally, this model successfully enables the visualization of pollution distribution on insulator surfaces in the field. The proposed method provides a basis for the non-destructive evaluation of transmission line insulators and offers solid support for the safe and stable operation of grid.