Wind Turbine Blade Icing Detection Method Based on SMOTE-SA-CNN

Detection of icing on wind turbine blades in wind farms is crucial for ensuring the safe and stable operation of wind turbines. The commonly used wind turbine blade icing detection algorithms based on deep learning often ignore the problems of data class imbalance in the wind turbine operation monitoring data and insufficient attention to icing feature parameters. Therefore, this paper proposes a wind turbine blade icing detection model based on SMOTE-SA-CNN.Firstly, the SMOTE (Synthetic Minority Over-sampling Technique) algorithm is employed to perform sample balancing processing on wind turbine monitoring data, effectively resolving the class imbalance problem caused by scarce icing condition data. Then, the self-attention (SA) mechanism is adopted to improve the model's attention to the key icing feature parameters. Finally, a convolutional neural network (CNN) is integrated to establish the blade icing detection model. Experimental results demonstrate that the proposed model achieves accuracy and recall rates of 96.48% and 97.86% respectively, providing reliable support for the safe operation of wind turbines.