Validation of Tipspeed models against field data¶
Tipspeed validates the layers of its digital twin and its analytics methods against field data. This page lists the results available to date.
In short: using pre-construction data only, the digital twin reproduces the wind at each turbine with an absolute bias of 0.11 m/s; yaw misalignment is estimated with a mean absolute error of 2.25° against nacelle lidar, in blind tests.
Validation results¶
| Model or method | Reference | Scope | Result | Publication |
|---|---|---|---|---|
| Wind flow | Masts, lidars and spinner anemometers, blind | 18 sensors on six wind farms, over 3 weeks to 1 year; pre-construction data only, from up to 11 km away and 1 to 11 years earlier | Bias −0.01 ± 0.10 m/s; absolute bias 0.11 ± 0.04 m/s; MAE 1.10 ± 0.17 m/s; correlation 0.81 ± 0.09 (per site, then averaged; wakes included) | Davoust, 2024a, 2024b; Davoust & Delaunay, 2025a |
| Yaw misalignment estimation | Nacelle-mounted lidar, blind | 23 tests, seven sites, two continents | 2.25° mean absolute error (updated results after the 2025 publication, on a larger validation dataset) | First results: Tipspeed, 2025b |
| Tuned turbulence and wind-profile corrections | SCADA power residuals | 11 sites, two of them offshore; 334 turbines; 1.55 million freestream, normal-operation timestamps | Correlation 0.78; bias below 0.15% of rated power | Davoust, 2026 |
| Wake models | Benchmark across wind farms | Nine wind farms | Mean absolute wake-loss error, relative to freestream energy: 5.3% TurbOPark Gaussian, 2.7% Park2, 2.4% Zong Gaussian, against 9.5% with no wake model | Davoust & Delaunay, 2026 |
Known limitations¶
- One yaw validation site showed a larger deviation, which is under investigation.
- The hybrid wake model achieved 1.7% in the same benchmark. It is a research topic and is not part of the delivered analyses.
- The optimal yaw offset is less reliable on some turbine technologies. It is an indicative metric, not a conclusive diagnosis.
Frequently asked questions¶
Have the Tipspeed models been validated offshore?¶
Two of the sites used for validation are offshore.
Who performed the validations?¶
Tipspeed performed the validations, on operational data from its projects. The wind speed and yaw misalignment tests were blind: the reference measurements were disclosed only after the estimates were delivered.
Blind test protocol¶
All turbines had previously been tested with a nacelle-mounted lidar. The lidar results were disclosed only after Tipspeed had produced its estimates. The tests covered a range of terrain complexities and turbine models, for several customers.