Abstract:

Aimed at the phenomenon of wind turbine blade icing, which is easy to occur in the cold areas, a method of icing detection of wind turbine blades using SCADA data was proposed. When the blades are icing, the power loss of generator will be increased, thus the method picks two variables, wind speed and power. Principal component analysis (PCA) was used to construct the projection feature on non-principal component direction which is sensitive to icing and active power of network. By choosing the optimal threshold, the logistic regression classifier is suitable for unbalanced classification. The effectiveness of this method was verified by the data of China Industrial Big Data Innovation Competition.

Key words: ice detection on wind turbine blade, SCADA data, non-principal component projection feature, optimal threshold selection, unbalanced classification