Hierarchical method for wind turbine prognosis using SCADA data

Abstract

Rapid development of wind energy requires effective wind turbine prognosis methods, which can give alarm before actual failure happens and hence enables condition‐based maintenance. A hierarchical method based on GP (Gaussian Processes) and PCA (Principal Component Analysis) is proposed in this paper for turbine prognosis using SCADA data. The method includes two levels of prognosis: 1) detect which wind turbine behaves abnormally and has potential defect; 2) determine the defective components in the abnormal turbine. On turbine level, the relationship between selected parameters and power generation is trained based on GP. Then the model residual, which is calculated as the difference between the estimated output and the actually measured power, can indicate whether the turbine is defective. On component level, the contribution of each SCADA variable to turbine abnormality can be given based on PCA method, and can be used for indicating the defective components. Field dataset including 24 failed turbines is used to validate the proposed hierarchical method. The validation results show that the proposed method can achieve wind turbine prognosis with 79% detection rate on turbine level and 76% detection rate on component level. Moreover, the method can provide several months ahead alarm before severe failure happens.