We take the site's own elevation at meter scale, mesh it, and run an AI turbulence model over that mesh,i with the operational forecast setting the conditions at the domain edges. Out comes a wind field with the ridge speed-up, the pass jet and the wake in it — at hub height, at each turbine, and in the time an AI responds rather than the hours a simulation would.
Hyperlocal
AI wind forecasts
for power markets
We resolve the wind over real ground at each machine, convert it to generation, and price it at your node
The approach
Three specialized AI models.
Simulation-grade wind at every turbine, at AI cost
Wind becomes power through a curve calibrated on the turbine's own observed generation rather than the manufacturer's sheet, plus a learned correction for the losses each machine imposes on the ones downwind of it, trained across multiple wind farms.
Predicted generation is carried through to the ERCOT nodal price at your settlement point and reported as revenue, over the same horizons as the wind and the generation, so the three line up hour for hour. Each hour arrives as a distribution rather than a single number, so a day-ahead offer is a choice of quantile — how much of the interval you are willing to have to buy back in real time — and not a guess at the middle.
Wind speeds up over the ridge and goes slack behind it. We resolve the ground instead, so every turbine gets the wind it actually sees.
We fit the curve to each machine's own generation rather than the manufacturer's sheet. That gap is what your forecast would otherwise carry: the sheet starts making power earlier than the machine does, so every hour on the rise comes in over-promised.
We run the model several times over to get a confidence region. That width is what your ERCOT bid is chosen against: bid a narrow hour at face value, and a wide one low enough to hold in real time.
What you receive
Numbers in your
systems, on a
schedule you can
plan against.
What arrives, and when
- Per turbine
- Hub-height wind speed and direction, generation in megawatt-hours, and revenue at your settlement point — for every machine on the site, not a site total.
- Refreshed
- Every ERCOT operational cycle (day-ahead)
- To start
- Turbine coordinates, hub heights and a settlement point. Historical generation is welcome and is what the validation is run against, but it is not a precondition: a site with no usable record still gets a forecast.
- Checked against
- SCED 60-day generation data, on your own site, before anything goes into production — and you see that error first.
How it runs
Trained on turbulence.
Adapted to your ground.
Tuned on your machines
once they have a record.
Pretrained on turbulence, adapted to your ground, tuned to your machines
What runs on every forecast cycle is a pretrained fluid AI model. It was trained on a large turbulence dataset, and it holds the accuracy of a full physics simulation at a small fraction of what it costs to run — which is the only reason a wind field this detailed can arrive on a forecast schedule at all. It is adapted to your real ground, with the operational forecast at the edges.
If your site has a record, we go further and tune the model on it — your own machines' generation, and whatever SCADA you have. We do this once your site is live and you want more out of it.