A novel pixel-level aging status evaluation method on on-site post insulators

In long-term operation, composite post insulators in ultra-high voltage (UHV) converter stations are prone to aging due to harsh environmental conditions such as high temperatures and intense UV radiation, leading to varying degrees of deterioration in hydrophobicity and mechanical strength of the insulators. This deterioration can subsequently trigger flashovers and power outages at the converter station, posing significant threats. Therefore, a precise and efficient detection method is proposed to effectively assess post insulators' aging status. Firstly, hyperspectral imaging (HSI) is employed to extract the spectral lines of post insulators in converter stations, and a simple and operable spray method is adopted as the standard for classifying the aging levels of the insulators. Secondly, spectral statistical characteristics (SSC) are proposed to reduce the dimensionality of hyperspectral lines, thereby improving computational efficiency. Subsequently, support vector machine (SVM), suitable for handling nonlinear hyperspectral data, is chosen as the classification model. Improved grey wolf optimization (IGWO) is proposed to obtain the optimal hyperparameters for SVM. Performance metrics such as overall accuracy (OA) and Kappa are utilized to evaluate the models, comparing them with five other commonly used classification models, including extreme learning machine (ELM), long short-term memory (LSTM), back propagation neural network (BPNN), random forest (RF), and traditional SVM. The results demonstrate that the SSC-IGWO-SVM model achieves the best classification performance, with an accuracy as high as 97.5%. Finally, this model is utilized to visualize the distribution of insulator aging status. The proposed method enables pixellevel accurate evaluation on post insulators in converter stations, providing crucial assurance for the safe and reliable operation of grids.