On the Selection of Input Variables for a Wind Turbine Load Monitoring System

Abstract The use of statistics from standard SCADA signals as inputs to monitor fatigue loads in wind turbines is a promising alternative to its continuous measurement with extra sensors and is the topic of this contribution. One of the fundamental challenges is the selection of input variables, a problem reflected by the various set of predictors used in previous investigations; a situation that hinders the generalization of the methodology. This work attempts to establish a single optimum set. First, it evaluates the impact of using different methods to select input variables on the accuracy of their predictions and then performs statistical tests on their residuals to establish a selection criterion between sets. It discusses the findings and relates the similarities and differences to the high cross-correlation found in the set of candidates. Finally, it drafts recommendations for the development of a robust system.