A Predictive Fault Diagnose Method of Wind Turbine Based on K-Means Clustering and Neural Networks

Maintenance costs can be greatly reduced by improving the prediction accuracy of wind turbine faults. In this paper, a new method combining K-means clustering and Neural Net classification methods are adopted to predict fault condition of wind turbine. First, the data preprocessing has been done with the K-means clustering of the data from SCADA (Supervisory Control And Data Acquisition) system. According to the internal association of condition data, by adopting the unsupervised learning method K-means, data with similar features are transformed to one cluster, after which a prediction with BP Neural Net classification method is made based on the converted clusters. Compared with the traditional Neural Net classification prediction methods, the proposed method mentioned above can improve the prediction accuracy by 3.5%, with which the abnormal state of the mechanical faults can be determined to a more accurate and timely degree.