4.1. Data source
The data used in this work were archived values from onshore MW class WTGs, from wind farms located in Canada containing more than 50 turbines. Values were measured at 1 Hz, but ten-minute averages were used in order to reduce the effects of outliers, noise and auto-correlation. Underperformance was simulated to allow us to measure the method's performance, i.e the time it takes to detect a shift of a given amplitude. Quality control of the data was accomplished by keeping only points were the WTG is available; producing power; and outliers were removed with a method inspired by the one used by Park et al. [12]. We define outliers as values that are not corresponding to normal operating conditions or if the WTG is not available over the whole ten minute period. Values resulting of an erroneous measurement or a database error are also removed. The purpose of removing these outliers is to increased the precision of the power curve and, thus, the reliability of the proposed monitoring method.