Posters

Abstract

A Wind Farm Digital Twin (WFDT) is a calibrated wind farm flow model that is continuously updated based on actual data from the wind farm. The objective of such methodology is to leverage wake information generated by the WFDT to better detect potential underperformance & wind speed measurement errors (anemometer bias, nacelle transfer function inaccuracies or other measurements artefacts). It also provides a very precise model of the wind farm production taking into account wake losses and wind turbines production status. It represents a major methodological innovation to quantify losses, especially from wakes, and assess the performance of a wind farm :  * Accurate temporal reconstruction of real operation : the digital twin reproduces, for every 10-minute time step, the exact operational state of each turbine (shutdowns, curtailments) as well as the incoming free wind conditions. Wake interactions are thus simulated within the real context of the farm at every moment. * Dynamic coupling with a calibrated wake simulator : a patented data-assimilation method based on a Kalman filter algorithm is applied to dynamically calibrate the wake model at each 10-minute timestep, ensuring that the model matches the local wind-speed deficits observed in the data for the current interval. The data-assimilation framework also adjusts the initially estimated free-stream wind speed to mitigate potential biases in the wind data (e.g. measurement biases, etc.). This calibration loop improves the representation of velocity deficits and ensures local and temporal consistency in the model. * Accurate estimation of the performance : since the digital twin accurately reproduces real conditions (free-stream wind, turbine states, local wind speed deficit), it allows the accurate quantification of the performance of the wind turbines, based on the simulated local wind speed and estimated power production. * Detailed and explanatory analysis : Beyond performance estimation, the WFDT offers a temporal reconstruction of the farm's actual operation with a calibration of wake effects. This makes it possible to precisely quantify the various production losses (curtailments, stops, etc.), especially wakes. This can have different applications such as Operational Yield Assessment, feedback on pre-construction yield assessment or accurate forecast of future production. The methodology has been applied to validation data of a wind farm from TotalEnergies portfolio. It has been demonstrated that : * The WFDT is able to predict the production of wind turbines with an average error of 1.6% over all wind turbines. * The WFDT is able to detect underperformance in the range of 1.5%-2% over 1 year of data, as well as a wind measurement error of 1% to 2%. On the same case, standard approaches (power curves and machine learning) are able to detect underperformance in the range of 2.5%-3%.  * The better detectability of the WFDT approach comes from the local wind information at each wind turbine that is provided by the calibrated wake simulator and which is of course not available for standard purely data-based approaches.