Comparison of novel SCADA data cleaning technique for wind turbine electric pitch system

Wind turbines typically do not operate in the ideal operating conditions, leading to abnormal behaviour that is reflected in their power curves. This abnormal behaviour can affect the performance of condition monitoring processes, as it may mask faulty behaviour. By cleaning other abnormal data, such as curtailment, models can learn the normal behaviour of the turbines. This paper presents a novel cleaning technique that utilises a combination of data binning and the Mahalanobis distance. This removes between 5 to 6% of the data, without great loss of normal data. When compared against other data cleaning techniques, the one presented in this paper produces a more ideal power curve. This technique could improve the performance of data-based condition monitoring techniques.