Transmission Line Ice Coating Prediction Using a Transformer-LSTM Hybrid Model

Research on predicting ice buildup on power lines is a key part of preventing and reducing damage in electrical systems. Currently, Long Short-Term Memory (LSTM) neural networks are widely used in this area because they work well with timebased data. However, LSTM struggles with long-term patterns. To address this issue, our study combines the Transformer model, which excels at capturing long-term dependencies, with LSTM to create a hybrid model. This combined approach takes the best features of both models, leading to better predictions for ice buildup. Tests under the same conditions show that the hybrid Transformer-LSTM model ($\mathrm{R}^{2}=0.9363$, MSE $=0.0417$, MAE $=0.1008$) works better than using just LSTM ($\mathrm{R}^{2}=0.9246$, MSE $=0.0493$, MAE $=0.1535$) or just the Transformer ($R^{2}=0.8365$, MSE $=\mathbf{0. 1 0 7 0}$, MAE $=\mathbf{0. 2 4 5 7}$). This proves that the combination of the Transformer and LSTM improves the accuracy of the prediction of ice on power lines.