SCADA based nonparametric models for condition monitoring of a wind turbine

Abstract

High operation and maintenance costs for offshore wind turbines push up the LCOE of offshore wind energy. Unscheduled maintenance due to unanticipated failures is the most prominent driver of the maintenance cost which reinforces the drive towards condition-based maintenance. SCADA based condition monitoring is a cost-effective approach where power curve used to assess the performance of a wind turbine. Such power curves are useful in identification of wind turbine abnormal behaviour. IEC standard 61400-12-1 outlines the guidelines for power curve modelling based on binning. However, establishing such a power curve takes considerable time and is far too slow to reflect changes in performance to be used directly for condition monitoring. To address this, data-driven, nonparametric models being used instead. Gaussian Process models and regression trees are commonly used nonlinear, nonparametric models useful in forecasting and prediction applications. In this paper, two nonparametric methods are proposed for power curve modelling. The Gaussian Process treated as the benchmark model, and a comparative analysis was undertaken using a Regression tree model; the advantages and limitations of each model will be outlined. The performance of these regression models is validated using readily available SCADA datasets from a healthy wind turbine operating under normal conditions.