False alarm detection in wind turbine by classification models

Wind energy is one of the most relevant renewable energy sources and it is expected to play a major role in the global energy production. The supervisory control and data acquisition system collects large datasets from wind turbines generating large amounts of alarms. These alarms are false in several cases, producing unnecessary downtimes and costs that affect the competitiveness of the industry. Operators are requiring new data analysis techniques to ensure the detection of false alarm and proper operation and maintenance tasks. It is presented a novel approach based on the analysis of coincident misclassification points produced by several machine learning algorithms for false alarms detection in wind turbines. The proposed algorithms are support vector machine, decision tree model, k-nearest neighbors, and ensemble trees. The use of this methodology demonstrates an increase in robustness and accuracy compared to the results of applying the algorithms individually. These methods classify intervals with alarms by analyzing the dataset of supervisory control and data acquisition system and validated by holdout. A real case study based on three real wind turbines is presented to test the methodology, and the results showed an accuracy of 99% for the classification algorithms. The analysis of the misclassified points detected 6, 15% of false alarms in the case study, indicating that this approach is promising to detect false alarms of wind turbines.