12 - Alarms and false-alarm analysis by support vector machine in wind turbines

Abstract

The operation and maintenance management are critical for the efficiency and competitiveness of the wind farms. The detection of false alarms is a significant variable for wind turbine maintenance. Large amounts of complex data are efficiently categorized by machine learning-based classifiers. The implementation of several support vector machine algorithms for the prediction and detection of false alarms in wind turbines is the novelty proposed in this research. A reliable tool for assessing the effectiveness of classification algorithms is K-Fold cross validation. The proposed methodology is verified by using Supervisory Control and Data Acquisition data from an actual wind turbine. The outcomes show a 98.6% accuracy rate for the quadratic support vector machine classifier. The analysis of the misclassifications obtained from the confusion matrix provides the necessary information, together with the alarm log and maintenance record, to determine whether it is a false alarm. The classifier can reduce the number of false alarms referred to as misclassifications by 25%. These results demonstrate that the suggested methodology is effective at identifying false alarms.