From sensor data to dispatch command, in one continuous loop.
Bluefield ingests NWP weather, live SCADA readings, and JEPX price signals every 30 minutes. The result is a rolling 24 to 48-hour battery charge-discharge schedule delivered to your EMS automatically.
A local correction layer on top of global NWP output.
Global weather models (JMA MSM, ECMWF, GFS) provide irradiance forecasts at 3 to 6 km grid resolution. For a rooftop or ground-mount solar site, that resolution misses the micro-meteorological effects that cause local cloud persistence and ramp variability.
Bluefield's gradient-boosted ensemble model ingests NWP irradiance, surface temperature, and humidity fields alongside the site's own historical SCADA generation. It learns the systematic bias between the NWP forecast and actual measured output for each site individually.
Output is a MW-by-30-minute solar generation curve covering the next 24 to 48 hours, updated every 30 minutes as new SCADA data arrives. The model automatically retrains on a rolling 90-day window to capture seasonal drift.
SoC-constrained optimization across the full planning horizon.
The dispatch optimizer takes the solar generation forecast and solves a constrained optimization problem: maximize settlement-window revenue while respecting battery state-of-charge limits, ramp rate constraints, and OCCTO interconnection scheduling rules.
Unlike pure arbitrage approaches that react to yesterday's price spreads, Bluefield's optimizer reasons across the full 48-hour horizon. It holds capacity in reserve when a ramp event is forecast, rather than discharging early on an average-price signal.
The output schedule is delivered as a JSON or CSV file to your EMS, updated every 30 minutes. You can inspect, override, or hold any time block before execution.
Forecast vs actual, every 30 minutes.
Every dispatch cycle, Bluefield records the forecast it issued, the schedule it sent, and the actual measured generation and SoC readings from your SCADA. The difference drives automatic recalibration of both the forecast model and the dispatch optimizer.
Your operations team sees a rolling accuracy dashboard: forecast error by hour-of-day, by season, and by weather regime. If a forecast quality threshold drops below the configured band, the system raises an alert before the dispatch window opens.
All data is stored in your own cloud tenancy. Bluefield never holds operational data outside your infrastructure.
Numbers we can defend, not numbers that sell.
Solar forecasting accuracy depends on site, season, cloud regime, and the quality of available NWP inputs. We do not publish headline accuracy percentages because a number calibrated on one site misleads operators at another.
What we can commit to: before contract, we run a backtested accuracy report on your site's own historical generation data. The report shows forecast error distribution by hour, by weather regime, and against the relevant JEPX settlement window. You see the actual numbers for your grid conditions before signing anything.
Illustrative ranges from backtests on western Japan grid-scale sites: hourly solar forecast MAE typically falls between 2 and 5 percent of installed capacity during clear-sky periods, and 6 to 12 percent during frontal cloud events. These figures are illustrative only and should not be taken as performance guarantees.
See the platform running on your grid data.
We run the backtesting report on your site's own generation history before you commit to anything.