Example forecast · pv-south-03 · 48 MW solarExample
04080120160MW-12hNow+12h+24h+36h+48h
P10 to P90P50Realized
P50 at noon36.1 MW
P10–P90 at noon29–41 MW
Clear-sky ratio0.82
Illustrative values, not a real site. Real forecasts carry their run time, model version and calibration window.

How it works

From read-only data to a decision you approve

Helivane is built to earn trust in steps. It forecasts in shadow first, then recommends, and acts only inside limits you have set and checked.

Process

Six steps, with timing

  1. Weeks 1–2

    Connect data

    Read-only exports of SCADA, metering and market data. We map assets and check data quality.

  2. Weeks 2–3

    Model the sites

    Terrain, layout, hub heights, panel geometry and power curves go into the downscaling and power models.

  3. Weeks 3–4

    Backtest

    We forecast your past 12–24 months as if live and compare skill and imbalance cost against your current forecasts.

  4. Weeks 5–8

    Forecast in shadow

    Live forecasts run next to your current ones. Nothing changes in operations.

  5. Weeks 9–12

    Recommend

    Dispatra proposes bids and setpoints with a reason. Your desk decides.

  6. Week 12

    Review

    We report skill, calibration and the money that decisions would have moved, against an agreed baseline.

Who does what

Responsibilities during a pilot

TaskYour teamHelivane
Data exports and accessProvide read-only exports or credentialsBuild and monitor the connectors
Site metadataConfirm layouts, curves and outagesBuild the site models
BaselineAgree the baseline forecast and cost methodCompute it the same way every day
Limits and approvalsSet limits and name approversEnforce them and log every action
DecisionsApprove or reject recommendationsExplain each one and record the outcome
ReviewJoin the week-12 reviewPrepare the report with all inputs
site.yaml
# site.yaml: one wind site in a Helivane pilot (example)
site: wf-north-07
type: wind
capacity_mw: 120
turbines: { count: 30, hub_height_m: 112, power_curve: vendor-4MW-v2 }
location: { lat: 53.912, lon: 8.871 }
data:
  scada: { source: historian-export, interval: 10min, history_months: 24 }
  metering: { source: settlement-files, interval: 15min }
market: { zone: DE-LU, settlement: 15min, gate_closure_min: 5 }
forecast: { horizon_h: 72, step: 15min, quantiles: [10, 50, 90] }
decisions:
  autonomy: L0          # recommend only
  max_bid_mw: 110
  approvers: [trading-desk]

Configuration

Each site is described in one file

A site file holds what the models need and the limits decisions must respect. It is versioned, so every forecast can be traced to the configuration that produced it.

  • Autonomy starts at L0 for every site
  • Limits live outside the model and are checked on every action
  • Changes are reviewed and logged

Outcomes

What a pilot gives you

The week-12 report is yours to keep whether or not you continue.

Skill

Forecast skill on your sites

Error and calibration by site, horizon and weather regime, next to your current forecast.

Money

Imbalance cost, recomputed

What the Helivane forecast would have cost in imbalance settlement, using the method you agreed.

Decisions

A log of recommendations

Every proposed bid or setpoint, its reason, your decision and what happened next.

Questions

About the process

How much history do you need?

Twelve months of 10-minute SCADA data is enough to start. Twenty-four months covers more weather regimes and gives a better backtest.

What if our data has gaps?

Gaps are normal. We flag them in the data-quality report and exclude affected periods from the skill comparison.

Who owns the models trained on our data?

Your site data and site models stay yours. Shared weather and physics models improve from everyone's error records, without exposing any customer's data.

Can we stop at the forecasting stage?

Yes. Many teams want Skycast alone. Decisions are an option, not a requirement.

Plan a pilot around your sites

Tell us your portfolio and markets, and we will propose a 12-week plan with a baseline you agree with.