Explainable Physical Deep-Learning Model for Overhead Transmission Line Icing-Thickness Prediction

Global warming has resulted in frequent extreme-weather events, among which frequent ice disasters significantly threaten the stability of power systems. As the complexity of prediction models increases, their accuracy and explainability must be ensured simultaneously. Therefore, we propose an explainable physical deep-learning model for overhead transmission line icing-thickness prediction. First, an optimization model is developed via Beluga Whale Optimization (BWO) method, which minimizes the prediction error. Next, the deep-learning prediction model is combined with physical and the Long Short-Term Memory (LSTM) models. The physical model considers physical laws, such as the wind deflection angle, wind load, and ice load. Additionally, we use Shapley Additive explanations to explain the effect of the input features on the output features and model prediction. Finally, the prediction model is validated using actual overhead transmission line icing-thickness data from 2022 to 2023. Experimental results indicate that the proposed model can effectively balance accuracy and explainability.