Wind Turbine Fault Identification in Sample Imbalance Scenarios Using FRBCS With GAN Oversampling and Metric Learning

The operation data from different wind turbines (WTs) is not universal, due to the influence of installation, location, and environmental conditions, which caused incomplete and imbalanced fault sample data, and influenced the fault identification accuracy. Therefore, this paper proposes a fuzzy rule-based classification system (FRBCS) based on generative adversarial network (GAN) oversampling and metric learning. The GAN oversampling algorithm is introduced to eliminate the imbalance of the actual samples, and the metric learning algorithm (MLA) is adopted to improve the separability of different fault samples of the FRBCS. 10 common faults of megawatt-level WTs are used to validate the effectiveness of the proposed method, the results show that the proposed method effectively improves fault identification accuracy in the presence of imbalanced fault samples.