ATR85 curtailment estimator
How the site curtailment estimator works — the engine behind the Solar Workbench.
What the estimator does
For a postcode, a capacity (MW), and a technology (solar, battery, or DC load), the estimator computes the expected number of curtailment hours per year, with a P10/P50/P90 confidence band across the available years. The result feeds the IRR comparison and is shown on the Workbench as the headline figure.
How the estimate is built
- Match the postcode to the grid. The postcode is resolved to the high-voltage substation that feeds it (the highest-voltage feeding point wins).
- Pull the history. Every 15-minute interval of modelled loading for that substation, bucketed by calendar year (UTC).
- Count threshold crossings. Every interval where modelled loading meets or exceeds the curtailment threshold counts. Thresholds, drawn from ATR85 practice:
- Solar / DC load: 95% (operators often curtail before a hard bind)
- Battery: 100% (only at a true bind)
- Percentile over years.P10 / P50 / P90 plus the worst year — linearly interpolated, with no Gaussian assumption.
- Wider band when history is short.With less than 12 months of data the P10/P90 band is widened linearly, up to ±50% at one month — so a few months of data is never presented as twelve years of certainty.
What the estimator explicitly does NOT do
- N-1 contingency. Lines are treated at their published thermal rating, with no outages. Real-world curtailment is often N-1: a neighbouring line goes out for maintenance, your connection jumps from 60% to 110%, and you get curtailed. This is not captured.
- Voltage / reactive power. The model ignores voltage and reactive power. In the Randstad, curtailment is often reactive-driven, so we under-detect there.
- Sub-substation detail. We resolve to the substation, not to your specific connection cable. If that cable is the bottleneck, we miss it.
- Operator policy. Operators may curtail ahead of the physics (precaution) or run past it (margin). We model the physics, not the policy.
Accuracy bands
Every estimate carries this band:
Classification-grade, not metering-grade. ±20–30 hours on a 100–300 h annual estimate in well-instrumented areas (FGU, Limburg); ±50–80 h in areas with heavy behind-the-meter solar (Brabant). Suitable for site screening and IRR comparison; not for bankable contract clauses.
The full accuracy-band table is on the main methodology page.