Prediction of wind turbine blades icing based on MBK-SMOTE and random forest in imbalanced data set
Yangming Ge, Dong Yue, Lei Chen · 2017
The icing problem of wind turbine blades caused by low temperature environment poses a serious threat to the power generation performance of wind turbines. The serious im-balanced real-time data of the wind turbine makes the traditional machine learning algorithm unable to make a good prediction for blades icing problem. To solve the problem, this paper proposes an improved SMOTE algorithm (MBK-SMOTE) which is based on the standard SMOTE algorithm. MBK-SMOTE makes samples distribution balanced by introducing the concept of sample's density and dividing the concentration area. Then we use the Random Forest algorithm to predict the icing event of wind turbine blades. Comparing the experiments among MBK-SMOTE+RF, standard SMOTE+RF and improved SMOTE+RF, we finally found that MBK-SMOTE+RF is more effective, and the prediction effect of MBK-SMOTE+RF is superior to other methods.