How a Solar Operator Cut Imbalance Costs with Power Forecast Intelligence

A large solar operator already ran a power forecast into its offtakers through Enlitia's data hub, using the same baseline provider it had relied on for years. To capture more of the market's value from the same megawatts, the operator partnered with Enlitia to build a Power Forecast Intelligence service — a layer that adjusts to what is actually happening on the plant, rather than a static curve from a single provider.

Challenge

A flat availability multiplier treats every inverter the same, but not every inverter has the same ceiling. Overequipped inverters clip and flatten against their AC rating around midday; others, sized close to 1:1, have no real ceiling at all. One flat percentage across a plant with both kinds gets the wrong answer either way. And relying on a single forecast provider means accepting whatever regime that provider is weakest in - no provider is best in every condition, every day of the year.

The goal was to compute availability at each inverter's own ceiling, feed that correction back into both the market offer and the provider's calibration, and add a second forecast provider without simply averaging the two.

What we built

A Power Forecast Intelligence service on top of the existing data hub and the operator's baseline provider - treated as one input among several, not a fixed starting point. Three pieces:

Real availability, per inverter's own ceiling - DC-to-AC ratio is a known property of each inverter, not a plant-wide assumption. A clipping inverter loses much less than a flat multiplier assumes, because surviving strings refill the gap; one with no headroom loses close to the full amount.
Recomputed forecast, closed-loop - regenerated on real availability, not rescaled from the provider's curve. The corrected number goes to market as the offer, and back to the forecast provider - baseline included - so its models calibrate on what actually happened, not on a number that was already wrong.
Provider Selector, not an Ensemble - a second provider sits alongside the baseline, and a selector picks the best forecast interval by interval based on the weather regime, rather than blending both into one averaged curve.

Results

5.3% real loss vs 10% declared - one clipping inverter's outage, flat-multiplier vs ceiling-aware. A neighboring inverter with no DC/AC headroom lost close to the full 10% for the same kind of outage - proof the two need different treatment.
Ensemble underperformed its own best but a Selector beat the best single provider at every site - validated on an 18-site portfolio: 8.77% NMAE vs 9.60% for the best single provider, 8.6% better, at all 18 of 18 sites.
EUR 44k-70k/yr selector upside on one plant on twelve months of real prices, moving from the baseline alone to a selector capturing 25-40% of the perfect-selection ceiling, against a full ceiling of about EUR 174,600/yr.

Impact across the organisation

The operator now scores its long-standing forecast provider on the same metrics as any new one, with nothing grandfathered in. Availability corrections became a two-way feedback signal, not a one-way adjustment before submission. And the choice between forecast sources became a live, interval-by-interval call instead of a one-time vendor decision.