Posters

Abstract

Modelling long-term energy losses due to wakes is a critical component of pre-construction energy assessments for wind farms. However, these models often carry uncertainty in long-term bias and exhibit significant temporal variability. Despite this, most asset owners lack insight into how operational wake losses compare to budgeted expectations, as well as the monthly and annual variability of these losses. This work introduces a scalable, data-driven algorithm that calculates a power wake loss time series using only 10-minute SCADA data, eliminating the need for prior wind flow modelling. The results can be aggregated by time period to determine monthly wake loss factors for budget reconciliation, or by sector and wind speed for wake model validation. Verification using open datasets from Kelmarsh and Penmanshiel wind farms reveals wake loss patterns consistent with industry expectations, while also highlighting significant monthly variability - a finding that can reduce uncertainty in operational yield reconciliation. The algorithm processes 10-minute SCADA data by cleaning and classifying turbine states (full performance, downrating, or stopped). Absolute wind direction measurements are calibrated using ERA5 data. For each timestamp, wake-free turbines are identified according to IEC 61400-12, with their measured power assigned as freestream power. For waked turbines, freestream power is estimated from the weighted average power of nearby wake-free turbines. Turbines in partial performance calculate lost power based on power curves, distinguishing it from derating losses. The resulting freestream power and measured power signals are used to derive wake production factors, which can be aggregated by sector, turbine, or time period for further analysis. When applied to the Kelmarsh and Penmanshiel datasets, the algorithm produced wake loss profiles aligned with expectations based on turbine layout. Notably, production gains were observed in sectors where flow bypasses upstream turbines. Monthly wake production loss factors revealed substantial variability: Kelmarsh exhibited an average loss of 8.9% with an interdecile range of 8.3%, while Penmanshiel showed an average loss of 6.1% with an interdecile range of 4.3%. This data-driven approach offers a scalable solution for estimating wake-induced production losses under any conditions. The results validate the method by demonstrating expected wake loss patterns and highlighting significant monthly variability, which is often overlooked in budget reconciliation and energy yield estimates. Beyond wind speed variability, this variability is influenced by the wind rose and turbines’ operational state - factors currently ignored in industry practice. Delegates will learn how variable wake losses impact monthly energy output and why this variability should be considered alongside wind speed variability. They will also discover the advantages of a scalable, data-driven approach for operational assets, reducing the cost and complexity of wake loss validation. Finally, those involved in wake model validation will gain insights into the potential of data-driven wake estimates for large-scale validation efforts.