Wind farm monitoring using Mahalanobis distance and fuzzy clustering

This paper proposes an approach for warnings and failures detection based on fuzzy clustering and the Mahalanobis distance. Both techniques are developed in a real wind farm for critical devices typically found in a wind turbine. A power curve is modelled using fuzzy clustering and parametric fitting techniques in a first step. Then, warnings and alarms recorded by a Supervisory Control and Data Acquisition system are analysed from their locations and distances to the curve. The Mahalanobis technique is selected for this purpose and its accuracy is validated with other methods considered. The research reveals the existence of zones with complex detectability for some winds speed and powers ranges. However, in contrast to a standard pattern, there will be differences in terms of distances. The usefulness of the findings lies in the inclusion of a real-time monitoring system applying easily available resources. The paper is understood as a complement to other specific and costly monitoring systems to ensure the implementation of actions before the occurrence of a failure. A large number of publications using the power curve can be found focusing on forecasting or market researches, but this trend is not usually extended to the wind turbine maintenance management.