The true future wind speed v(t) is a pre-realized Ornstein–Uhlenbeck (mean-reverting) stochastic path — the standard textbook model for short-term wind fluctuation around a slowly-varying mean:
dv = θ(μ − v)·dt + σ·dW
θ = reversion rate (how fast gusts decay)
μ = mean wind speed (slider)
σ = turbulence intensity × μ (slider)
dW = Gaussian white noise
Each turbine converts wind speed into electrical power through the standard three-region power curve used in real turbine datasheets:
P(v) = 0 for v < v_cut-in (≈3.5 m/s)
P(v) = ½ρA·Cp·v³ for v_cut-in ≤ v < v_rated (≈12.5 m/s)
P(v) = P_rated for v_rated ≤ v < v_cut-out (≈25 m/s)
P(v) = 0 for v ≥ v_cut-out (safety shutdown)
The forecast wind speed at lead time h combines the same persistence-decay toward the mean used by real short-horizon forecasters, plus an explicit forecast-uncertainty term modeled as a random walk whose variance grows linearly with lead time (so its standard deviation grows as √h) — the standard way forecast-skill loss is characterized statistically:
v_forecast(t+h) = μ + (v(t) − μ)·e^(−h/τ) + N(0, k²·h)
k = forecast error growth rate (slider)
Because power depends on the cube of wind speed, a small relative wind-speed forecast error dv/v translates into a roughly three times larger relative power-forecast error near the steep cubic part of the curve: dP/P ≈ 3·dv/v. The "Error amplification" readout is the live-measured ratio of relative power error to relative wind error — it should sit clearly above 1 whenever the true wind sits in the cubic region, and collapses toward 1 (or below) once the turbine is pinned at rated power, where dP/dv ≈ 0.