Definition
Twin-predicted structural loads, energy production, and thermal behavior compared against sensor-measured actuals, segmented by operating regime and reported at the worst-performing regime rather than the average. The regime where the twin is weakest is the one that determines whether it can be trusted.
The Measurement Gap
Digital twins in renewable energy operations serve a specific high-stakes purpose: they simulate structural loads, energy output, and thermal behavior to support decisions that must be made faster than physical inspection allows. During a storm event, the digital twin tells the asset manager whether the turbine is approaching structural load limits that require curtailment or shutdown. During normal operations, the twin feeds lifetime projections that drive capital planning decisions years into the future.
The measurement gap is that twin accuracy is almost never evaluated by operating regime. An aggregate fidelity score of 92% might mean the twin is 96% accurate during normal wind conditions (which occur 80% of the time) and 72% accurate during extreme gust events (which occur 5% of the time but carry 90% of the structural risk). The aggregate looks acceptable. The extreme-regime accuracy is dangerously inadequate for the decisions it supports.
This creates what TRFI calls the "trust cliff": the gap between normal-regime fidelity and extreme-regime fidelity. The larger the trust cliff, the more the organization is relying on the twin precisely where it performs worst. Curtailment decisions during storms, load protection during icing events, and thermal management during heat waves all consume twin outputs in extreme regimes where the twin was likely trained on insufficient data.
The decision TRFI supports is whether the asset manager can trust the digital twin's structural load estimate during extreme events to decide whether curtailment for protection is warranted, or whether the twin is least accurate exactly when it matters most and requires human override.
Formula
Compare twin-predicted outputs (structural loads, energy production, thermal behavior) against sensor-measured actuals, segmented by operating regime. Compute the normalized error per regime. The regime with the highest error is the regime where trust should be lowest.
Formula (Notation):
TRFI(r) = 1 minus (1/N_r) times the sum of |Twin_Output_i minus Measured_i| / Measured_i, for all periods i in regime r
Variable Definitions:
| Variable | Definition |
|---|---|
| r | operating regime (normal, high wind, extreme gust, icing, high irradiance, sandstorm, or other site-specific regimes) |
| Twin_Output_i | twin-predicted value for period i. This may be structural load in kN, energy output in kWh, or temperature in degrees C, depending on which twin output is being validated |
| Measured_i | sensor-measured actual value for the same parameter and period |
| N_r | number of comparison points in regime r |
Edge Cases and Measurement Boundaries:
Extreme regimes accumulate few data points by definition. Accumulate comparison data over multiple seasons before computing TRFI for rare regimes, and flag any regime with fewer than 15 comparison points as "low confidence." Reset the TRFI baseline after each twin recalibration event to avoid mixing pre- and post-calibration accuracy. When the twin produces multiple output types, compute TRFI separately for each (loads, energy, thermal) because accuracy may vary significantly by output type.
Confidence Grading (required alongside every TRFI value):
| Grade | Basis | Interpretation |
|---|---|---|
| A: Empirical | 5+ observed events in measurement window provide direct tail observation | Measurement grounded in observed outcomes; highest confidence |
| B: Hybrid | 1-4 observed events plus Monte Carlo simulation with Bayesian updating | Blends observation and simulation; moderate confidence |
| C: Simulated | Zero observed events; derived entirely from Monte Carlo simulation | Modeled, not observed; treat as indicative, not validated |
A TRFI of any value at Grade C is a simulation that agrees with itself. It tells you the model is internally consistent but not that it matches reality. Grade C scores should trigger proactive stress-testing against historical near-miss events and cross-asset analogues rather than acceptance as calibrated.
Proposed formula, pending validation pilots.