Data-driven predictive analytics of unexpected wind turbine shut-downs

In this study, a novel data-driven framework is proposed to offer predictive analytics of wind turbine (WT) unexpected shut-downs based on data collected by the supervisory control and data acquisition (SCADA) system. A new parameter, the remaining functional life (RFL), is introduced to describe the length of a period until the next WT shut-down and a binary target parameter is created based on the RFL for indicating impending unexpected WT shut-downs. A two-stage data-driven framework is proposed to develop the predictive analytics model of the unexpected WT shut-downs based on SCADA data. The first stage employs clustering methods to automatically cluster WT SCADA data through unsupervised learning. In the second stage, based on clusters of SCADA data, famous classification methods are applied to develop models for inferring the binary target parameter. To validate the proposed data-driven framework, case studies and intensive computational experiments are conducted. Computational results confirm that meaningful predictive analytics of unexpected WT shut-downs can be produced through the proposed data-driven framework.