OCF Solar GB Webinar: AI Solar Forecasting for the Grid

Why solar forecasting matters for the GB grid

Great Britain has roughly 24GW of installed solar capacity, and ambitious targets mean that figure will keep climbing over the next decade. That's great news for decarbonisation, but it creates a new challenge for grid balancing.

Solar generation is inherently uncertain. To cover that uncertainty, grid operators hold energy reserves on standby, often in the form of gas turbines that run inefficiently when they aren't operating at full capacity. Better solar forecasts mean less uncertainty, which means less reserve, lower costs and lower emissions. Open Climate Fix estimates that good forecasting could help eliminate up to 50 million tonnes of CO2 emissions a year.

How OCF Solar works

OCF Solar is a short-term solar PV forecast focused on intraday and day-ahead horizons. It runs every 30 minutes and forecasts each of GB's half-hourly settlement periods.

Built from the grid up

Rather than forecasting GB as a single block, OCF Solar models each of the roughly 317 Grid Supply Point (GSP) regions individually. A separate machine learning model then combines these regional forecasts into a national view, which delivers more accuracy than simply adding them together.

The forecast is trained against PV_Live, the solar generation outturn dataset from the University of Sheffield. The updated PV_Live outturn is based on around 25,000 real household solar sites, extrapolated using capacity estimates for each GSP region.

Multimodal machine learning

OCF Solar combines several data sources:

Numerical weather predictions from the Met Office's 2km UKV model and ECMWF, including cloud cover, visibility, fog, snow, temperature and wind speed
Satellite imagery from geostationary satellites covering Europe, updated every five minutes
Around five years of historical forecasts to train on

For each region, the model sees a wide crop of satellite imagery, so it can spot clouds that are about to move in, alongside crops of weather data. These inputs pass through PVNet, Open Climate Fix's deep learning architecture, to produce a probabilistic forecast.

The multimodal approach is where the edge comes from. Satellite-only models perform strongly in the first hour or so but lose skill quickly. Weather-model-only forecasts are flatter and more consistent across the horizon. Combining everything gives OCF Solar the best of both: the short-term uplift from satellite data and the stability of numerical weather predictions further out.

How accurate is it?

Independent benchmarking by industry partners has shown that OCF Solar consistently outperforms other solar forecasts currently on the market.

Inside the OCF Solar platform

The live demo covered the key features of the OCF Solar user interface:

National and regional forecasts: an interactive map of every GSP region, with individual forecasts for each
Probabilistic forecasts: quantile bands alongside the expected generation, so users can quantify uncertainty. User feedback on the quantiles has been especially positive
PV_Live comparison: the live intraday estimate and the updated outturn plotted against the forecast
Forecast history: see how forecasts made one, two or four hours ahead compared with what happened
Single-source models: separate satellite-only, Met Office-only and ECMWF-only forecasts that help explain why the main model is behaving as it is
Seasonal mean and quantiles: a view of how today's generation compares with the past five years for this time of year, scaled to today's capacity
Constraint boundaries: select multiple GSPs to see aggregated solar generation behind a grid constraint
Satellite imagery in the UI: a brand-new feature showing visible, infrared and water vapour channels over time
CSV download: export forecasts for your own internal tools

The OCF Solar API

Everything in the UI is also available via API, including national and GSP-level forecasts, probabilistic levels and single-source models. Users can pull a full year of historical forecasts for testing and benchmarking. A new API is launching very soon, and the webinar includes a sneak peek.

Real-world impact: NESO and beyond

OCF Solar began as an innovation project with NESO and is now used in the NESO control room for real-time balancing decisions. An independent report on Dynamic reserve setting, found that OCF Solar was saving NESO around 100MW of reserve per settlement period. That equates to roughly 300,000 tonnes of carbon savings and around £30 million in cost savings

Open Climate Fix has also applied OCF Solar to distribution networks. With UK Power Networks, we used primary substation power flows to estimate embedded solar capacity across highly localised regions, helping unlock flexibility. A similar project with National Grid Electricity Distribution is under way, forecasting demand, solar and wind to improve network operation and reduce blind spots.

Beyond network operators, OCF Solar is used by energy trading houses of all sizes and by smart home technology companies.

Frequently asked questions

What is OCF Solar? OCF Solar is an AI-powered solar PV generation forecast for Great Britain, developed by the non-profit Open Climate Fix and used by NESO.

How often is the forecast updated? Every 30 minutes, covering GB's half-hourly settlement periods across intraday and day-ahead horizons.

Does OCF Solar provide regional forecasts? Yes. It forecasts each of GB's roughly 317 Grid Supply Point regions as well as the national total.

Can I access the forecast via API? Yes. Everything in the UI is available via API, including a year of historical forecasts.

Can I trial OCF Solar? Yes. Trials and evaluation data, including five years of backtests, are available on request.

Try OCF Solar for GB

Want to evaluate OCF Solar for your trading, grid or energy business? Get in touch with Zak to arrange a trial or access historical forecasts and backtests for your use case.

Book a Demo