Prediction of Discontinuous Time Icing Events Based on XGBoost-GRU Model

As the global transition to sustainable and renewable energy sources accelerates, wind energy is becoming increasingly vital. However, in colder regions, one of the key operational challenges for wind turbines is ice accumulation on the blades. This study presents a hybrid model that combines XGBoost for feature extraction with a Gated Recurrent Unit (GRU) network for time-series prediction. By analyzing the last 7 minutes of operational data, the GRU model forecasts icing conditions at specific future intervals. This method not only detects the onset of icing but also provides operators with critical information about the timing of such events, enabling early preventive measures. In the hybrid framework, XGBoost enhances the accuracy of predictions by identifying and selecting the most relevant features. The data for this study is sourced from a wind farm located in Shanxi Province, China <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(113^{\circ} \mathrm{E}, 37^{\circ} \mathrm{N})$</tex>.