A universal modeling approach for wind turbine condition monitoring based on SCADA data

An universal approach for building the wind turbines (WT) prediction models has been proposed. BPNN is used to establish the forecasting model from the perspective of large data analysis based on the SCADA data. First, the power quality of wind turbines can affect the power supply to the grid, thus the power can represent the performance of the WTs. So the active power is selected as the output of the model. Other condition parameters are clustered based on k-means methods in which the silhouette coefficient is used as the basis of clustering. Second, correlations within the class are calculated and unrelated quantities in each class are maintained to reduce the dimensionality of the data. Then the remaining condition parameters associated with the output power are retained. Thus the large condition parameters can be represented by several representative parameters as input to the prediction model. Finally, according to the relative error probability distribution to monitor WTs' condition, this method can be widely used in any wind farm. The proposed method can effectively remove the redundant parameters, thus can be more intuitively and quickly used for condition monitoring.