Presentations

Abstract

Wake-steering tools are now widely accessible through open-source software, supported by active user communities. Yet the key challenge for broad industry adoption remains field validation: deploying a Wake-Steering Controller (WSC) on an operating wind farm and demonstrating that predicted energy gains are realised under real, time-varying conditions. Moreover, different engineering wake models, and different parameter choices within the same model, can yield materially different gain estimates. We present a SCADA-data-driven workflow for model selection, data-driven tuning, and performance validation in FLORIS (FLOw Redirection and Induction in Steady State), with an emphasis on improving wake-steering predictions across sites and supporting applicability to day-to-day operational decision-making. Our case study evaluates approximately one year of wind farm SCADA and WSC time-series data from an operational toggle test, with a 50-minute alternation between baseline operation (toggle-off) and wake-steering control (toggle-on). Two upstream turbines apply yaw-based wake steering to benefit downstream turbines. Only valid operational data from the toggle test are retained. Details about best-practice data evaluation will be presented in the workshop. Specifically, the transition effects between consecutive toggle states are mitigated through consistent filtering rules, and the assessment is restricted to periods when turbines are available in full performance so that internal wake interactions remain consistent throughout the evaluation period. The filtered dataset is then used to determine energy production during toggle-on and toggle-off periods. Energy performance is quantified using an energy-gain ratio derived from reference turbines, enabling fair comparison between toggle-on and toggle-off samples while reducing sensitivity to natural variability in wind speed, turbulence, and seasonality, enabling fair comparison between toggle-on and toggle-off samples. Statistical uncertainty is reported via bootstrap confidence intervals computed from the filtered dataset, and sensitivity checks are performed for wind-direction binning choices and reference-turbine selection rules. On the modelling side, we benchmark multiple FLORIS steady-state wake models, including Jensen/Park, Gauss/GCH, Cumulative Curl, TurbOPark, and Empirical Gaussian formulations. Key parameters (for example, wake recovery and expansion settings, and yaw-power loss representation) are tuned via data-driven optimization technique against baseline wake losses and the measured incremental effect of wake steering. The preferred model-parameter combination must reproduce both baseline and controlled performance, within the measured uncertainty, without overfitting a single wind sector. The resulting ranked model set is used to quantify how much prediction spread remains after tuning and highlights dominant drivers of model-data mismatch that matter in practice, such as wind-direction bias, yaw-actuator dynamics, etc. The workshop presentation will share the findings from the validation of wake-steering performance based on one year of wind farm SCADA and WSC time-series data from an operational toggle test and will propose best-practice recommendations to enable comparison across sites and different steady-state wake models, helping operators, OEMs, and investors de-risk wake-steering deployment.