Intelligent identification and prediction of wind turbine icing status by integrating wind chill temperature features and ensemble learning

Icing of wind turbines in mountainous wind farms during winter poses a serious threat to the safe operation of power systems. However, due to the constraints of complex terrain and limited monitoring conditions, the absence of multidimensional observation data makes intelligent identification and forecasting of icing particularly challenging. To address this issue, this study proposes a wind turbine icing identification and prediction approach tailored for mountainous wind farms with limited data availability. A wind chill temperature index is innovatively introduced to enhance meteorological feature representation, and a Stacking ensemble model integrating Random Forest, XGBoost, Long Short-Term Memory, and Transformer is developed. Experimental results demonstrate that the proposed method achieves an identification accuracy exceeding 92.3% using only temperature and wind speed data, improving by more than 2.8% compared with the best individual model. Furthermore, validation with a cold wave event driven by ECMWF IFS forecast data confirms the model’s ability to predict turbine icing four to five days in advance, with accuracies of 85% and 70%, respectively. These findings provide a reliable technical solution for icing warning in mountainous wind farms and highlight significant engineering application potential. 冬季山地风电场风机覆冰对电力系统的安全稳定运行构成严重威胁.针对监测数据不足导致的识别与预测困难, 本文提出一种融合风寒温度特征与集成学习的风机覆冰智能识别与预测方法.研究创新性地引入了风寒温度特征以增强气象特征表达, 并构建融合随机森林,XGBoost,LSTM与Transformer的Stacking集成模型.结果表明, 该方法仅基于温度与风速即可实现92.3%以上的识别精度, 较最佳单模型提升超过2.8%.一次寒潮事件验证显示, 该模型能够提前4至5天准确预测风机覆冰, 预测精度分别达到85%和70%.