Cost-Sensitive Extremely Randomized Trees Algorithm for Online Fault Detection of Wind Turbine Generators

The number of normal samples wind turbine generators is much larger than the fault samples. To solve problem imbalanced classification in generator detection, a cost-sensitive extremely randomized trees (CS-ERT) algorithm proposed this paper, which learning method introduced into an (ERT) algorithm. Based on misclassification cost and class distribution, gain (MCG) as score measure CS-ERT model growth process to improve accuracy minority classes. Hilbert-Schmidt independence criterion lasso (HSICLasso) feature selection used select strongly correlated non-redundant features doubly-fed generators. effectiveness was verified by experiments four different failure datasets experiment results show that average missing detection rate, gMean improved better those ERT In addition, compared with CSForest, AdaCost MetaCost methods, has real-time performance.