Power curve

A power curve maps wind speed to electrical output: nothing below cut-in, a steep rise through the middle of the range, flat at rated power, and zero above cut-out. Manufacturers publish one per turbine model, measured under reference conditions defined by the IEC 61400-12 standard.

The steep section is where accuracy matters most. Available power in the wind rises with the cube of wind speed, so in the middle of the range a small error in the measured speed produces a large error in the expected power. That sensitivity is why the nacelle anemometer and its calibration are part of the measurement problem rather than a detail of it.

The published curve is a specification rather than a description of your machine. Real turbines deviate from it through blade soiling and erosion, control settings, air density, terrain, and wake effects from neighbours - and the deviation grows with age. Judging a ten-year-old turbine against a datasheet curve therefore produces a number that is part equipment condition and part everything else, which is not actionable.

The alternative is a learned curve: build each turbine's expected output from its own operating history under matched conditions, then measure the live machine against itself. Underperformance becomes visible as a gap against a baseline that already accounts for where the turbine stands and what it has been through, rather than against an idealized machine that does not exist on your site. That learned baseline is what a normal behavior model generalizes to every other channel the turbine records.