Abstract: The prediction of icing on power transmission lines is a key technology to ensure the safe operation of the power grid. Icing prediction is a complex task characterized by high-dimensional nonlinearity and multimodal heterogeneity, as it necessitates the comprehensive consideration of terrain and meteorological changes. This paper proposes a deep learning approach based on an ensemble random configuration network to predict icing on transmission lines. Icing transmission line recognition is enhanced by utilizing multiscale fusion of wavelet mod-maxima for icing image edge detection. Considering features such as micro-geography and micro-meteorology in historical observational data, a Boosting ensemble learning framework is employed along with a random configuration network prediction model to forecast icing conditions on transmission lines. Case study analysis demonstrates that the proposed ensemble model outperforms individual models, effectively achieving icing transmission line recognition and thickness prediction, thereby enhancing model generalization capability and improving the accuracy of icing disaster prediction.

Key words: icing prediction, stochastic configuration network, ensemble learning, prediction model, multi-scale fusion