Abstract

In this paper, we introduce a data-driven framework which combines a feature derivation to be the backbone of a behaviour model developed for fleet of wind turbines. The feature derivation component involves the use of a multivariate feature selection algorithm based on a novel Combined Power Predictive Score (CPPS), where the information content of combinations of variables is considered for the prediction of one or more key parameters. A community detection algorithm, based on Complex Network Analysis, is then used to unveil groups of wind turbine (within a fleet) featuring similar behaviour. The algorithm was tested on about 50 wind turbines, including diverse OEMs. The proposed algorithm employs the XGBoost regressor considering lagged time series to compute the Combined Predictive Power Score, and the results demonstrate the flexibility of the multi-input multi-output formulation, outperforming the standard monovariate formulation. The framework is used on active power target and it demonstrates to be a viable option in normal behaviour modelling derivation as it permits to isolate communities of wind turbines based on their actual operational history. This circumstance promised to be the forerunner of asset-specific O&M strategies.

This content is only available via PDF.
You can access this article if you purchase or spend a download.