Multi-Fault Classification of Unbalanced Wind Turbine Scada Data Using Stacking Classifier

Wind turbines have a rapidly increasing penetration worldwide as a clean and renewable energy source. The main factors and risks affecting the operational costs in wind farms established in large power plants are turbine failures and long downtimes. Periodic maintenance of wind turbines can boost efficiency and reliability. However, traditional routine checks make it difficult to detect wind turbine failures. Fault detection through a supervisory control and data acquisition (SCADA) system depends on the reliability of its data and the success of the model. This paper proposes a new model to classify the failures of the wind turbine condition monitoring system. In order to improve the success of the model, historical turbine measurement and failure data are pre-processed and filtered. Then, class balancing and data augmentation methods are applied with the Synthetic Minority Oversampling Technique (SMOTE). In the model; CatBoost based Stacking Ensemble classifier is proposed which includes XGBoost, LightGBM and Random Forest classifiers. This stacking ensemble classifier consolidates each classification model to improve the accuracy. The output of these models is used as input to the meta classifier of Catboost. The study shows that the proposed method diagnoses each fault successfully.