Automated Underperformance Detection for Wind Turbines

Publieke samenvatting / Public summary


The AUDIT project brought together the utility company Fortum and research institute TNO to investigate a novel approach to performance monitoring for wind farms, which we called social statistics.

Short description

The developed framework allows online detection of calibration and control errors, by allowing turbines to compare their performance with each other. Rather than treating each turbine as a separate entity—which results in uncertainties in instrumentation often dominating any underperformance issues which are being investigated—social statistics enables fast, online inference of disagreements, and their possible causes. A combination of domain knowledge and time series statistical methods are used for this. This framework was implemented in Python and—together with algorithms targeting turbine yaw misalignment (which causes loss of energy production)—applied to a long time series of onshore wind farm data provided by Fortum. This time series comprised turbine sensor data in the 10-minute-averaged format typically stored by wind farm owners. Additionally, several campaigns of nacelle-mounted Lidar were provided, which directly detect turbine yaw angle misalignments.


It has been found that sensor calibration (in particular for nacelle orientation) is a concern in this wind farm, and through discussion is it found to be a common issue. Yaw sensor miscalibration and turbine misalignment with the wind are often indistinguishable, and an external source of truth is required to separate them (as the effect on power output is difficult to distinguish with low-frequency data from other natural and mechanical causes of energy loss). Nevertheless, a few algorithms have been implemented in this project—with and without the requirement of installing Lidar—resulting in a prototype system which can produce useful displays and alerts for the wind farm operator (see example below). Temporary online data correction is also possible, until maintenance can be conducted.