We forecast GB electricity demand — a week to two years ahead — including the peak and off-peak split that power buyers hedge against. Built entirely from public data.
95.1% accurate
forecasting 2 years ahead
96.13% accurate
forecasting 12 months ahead
96.71% accurate
forecasting 1 week ahead
18 public feeds
all public — nothing proprietary
What these terms mean — plain English, no jargon
MAPE
Mean absolute percentage error — on average, how many percent the forecast is off by. Lower is better: a 4% error means the forecast is within ~4% of actual, i.e. ~96% accurate.
Lead time
How far ahead the forecast is made (a 3-day lead = forecasting three days into the future).
MW / GW
Megawatts and gigawatts — how electricity demand is measured. 1 GW = 1,000 MW; GB demand runs roughly 20–45 GW.
Where demand is now — GB by distribution region (estimate, coloured by modelled demand)
Tap or hover a region
24,634 MW national
14 GB DNO licence areas · 2026-10-09 · coloured by modelled demand. Select any region for its share, population and local temperature.
Ranked by modelled demand
South East England3,011 MW
London2,970 MW
North West England2,534 MW
Southern England2,511 MW
East England2,097 MW
West Midlands2,015 MW
Yorkshire1,853 MW
East Midlands1,674 MW
South West England1,457 MW
South and Central Scotland1,358 MW
North Wales, Merseyside and Cheshire1,021 MW
North East England932 MW
South Wales712 MW
North Scotland489 MW
lowerhigher demand
National is measured; regional is a clearly-labelled estimate. ESTIMATE: national demand split by real DNO-area population (ONS) and adjusted by each area's real temperature. No public multi-year half-hourly regional demand history exists to backtest against — national is the rigorous, backtested part. National demand is split across the real distribution regions by population and adjusted by each area’s real temperature, because no public regional demand history exists to measure against. The 14 regions are Britain’s real distribution licence areas. Hover or tap a region for its modelled figure.
The forecast — choose a horizon (recent actuals + the road ahead, with widening uncertainty)
recent real demand Gridcast forecast P10–P90 band
96.71% accurate
forecasting 1–7 days ahead on the live weather forecast (3.29% average error)
+47%
more accurate than the standard benchmark (6.24% error)
28.1GW
Peak forecast demand by 14 days (2026-10-19)
±4.3%
Measured uncertainty band (P10–P90 of realised error)
The near-term forecast. Out to about two weeks the model uses the live weather forecast — the sharpest, most trustworthy view.
~95% accurate
forecasting 30 days ahead — 4.70% average error
+23%
more accurate than the standard benchmark (6.13% error)
29.4GW
Peak forecast demand by 30 days (2026-11-05)
±7.9%
Measured uncertainty band (P10–P90 of realised error)
Beyond ~2 weeks weather is not predictable, so the 30-day view uses typical seasonal weather. The error rises a little versus 14-day — that rise is exactly the value of having a real weather forecast.
~95% accurate
forecasting 90 days ahead — 4.76% average error
+22%
more accurate than the standard benchmark (6.10% error)
33.7GW
Peak forecast demand by 90 days (2027-01-07)
±8.1%
Measured uncertainty band (P10–P90 of realised error)
A seasonal forecast on typical weather. Useful for planning the shape of the season, not a day-precise number — the uncertainty band is wider and shown in full.
~95% accurate
forecasting 365 days ahead — 4.86% average error
+22%
more accurate than the standard benchmark (6.25% error)
34.2GW
Peak forecast demand by 365 days (2027-01-14)
±8.6%
Measured uncertainty band (P10–P90 of realised error)
A full-year outlook on typical seasonal weather — a forecast of the annual shape, explicitly NOT a weather-driven one. Treat the band, not the line, as the answer.
The dark line is the latest real demand; the teal line is the model’s forward forecast for the selected horizon. The next 16 days use a live weather forecast; beyond the weather forecast’s reach the line falls back to typical seasonal temperature. The shaded band is the measured P10–P90 error range, not a hand-drawn guess — tightest on days with a live weather forecast, wider on the long-horizon tail.
How accurate is it?
We forecast each day using only the weather forecast that was available at the time, then check it against the demand that actually happened.
3.29% erroraverage error forecasting a week ahead — about 97% of actual demand, +47% more accurate than the standard benchmark, across 4,844 forecasts.
What we predicted vs what actually happened — day by day
Every day from 1 Jul 2024 to 23 May 2026, forecasting 3 days ahead: our forecast against the demand that actually happened. Where you can see teal, the two briefly diverged; everywhere else they sit on top of each other.
Our forecast Actual demand gap between the two
692 consecutive days, each forecast 3 days ahead and scored against what actually happened.
And it runs live. We log every forecast the day we make it and score it against what actually happens — a public, dated record that cannot be back-fitted. Most recent: on 2026-10-08 we forecast 25,261 MW for 2026-10-09; actual demand was 24,634 MW — 2.5% out.
How far ahead can it carry? (the long-horizon question, measured)
Suppliers and large buyers hedge power forward — months to a few years before it is used — so the question that matters is how far ahead a demand forecast still holds. Measured on monthly GB demand, 2013-01 to 2026-04 (160 months), against what actually happened.
95.10% accurateforecasting the monthly total a full 24 months ahead (4.90% error) — and 96.3% at 1 month. Across the whole 1–24-month range the error stays in a narrow 3.6–4.9% band (634 forecasts).
We tested it against the established statistical methods professional forecasters rely on — across 160 months of real demand.
Demand this far ahead is driven by predictable patterns rather than the weather, which cannot be forecast beyond about two weeks — that is why a 12-month view holds.
Every forecast comes with a measured range. Across these horizons the 80% range held the actual demand 80% of the time and the 95% range 93% — the ranges are sized from real outcomes, not assumed.
The shape of the day (peak vs off-peak)
Power is bought in blocks — peak (07:00–19:00) and off-peak — and the split between them is shifting under everyone’s feet. Rooftop solar has carved out the middle of the summer day: the daytime peak used to sit ~30% above the overnight base; today it is ~3%. We measure that, and we forecast where the split goes.
What has happened — the daytime premium, 2013–2025
95.6% accurateforecasting the peak block a year ahead (94.6% even two years out). Modelling the rooftop-solar shift makes our peak-block forecast significantly more accurate than a standard seasonal model that sees only demand net of solar.
The forecast split — pick a horizon
Our forward forecast of each block’s demand. Drag the horizon, or read the chart: the deep winter peak, and the summer months where peak demand now sits below the overnight base — the solar effect, projected forward.
—
For anyone hedging peak and off-peak blocks, the split is the trade — and its drift is a structural shift, not weather. That is exactly what a months-to-years forecast should carry, and what we built.
Why this matters for energy trading (the localised-demand angle)
Demand is the predictable side of the power market.
Supply (wind, solar, interconnectors) is volatile; demand is driven by weather and the calendar — which is why a 97%-accurate national forecast is tradeable signal, not noise.
A sharper demand call sizes the residual — a forecast 47% more accurate than the benchmark on the demand leg tightens the whole net-load estimate.
Localised swings are where the dislocations live — the regional view surfaces demand shifting between regions, not just in total.
Auditability is the edge — a forecast measured nationally and labelled where it is an estimate is one you can size a position against.
Illustrative of the use-case, not a trading recommendation. Gridcast forecasts demand only; it does not forecast price.
What the edge is worth (we measured the volume; you set the price)
A demand forecast earns its keep by sizing a hedge better. We measured one quantity — the volume of national mis-hedge our forecast avoids versus the seasonal carry a buyer falls back on at 3–12 months. Set your own forward–outturn spread and share of load; the value follows. We forecast demand, not price.
3,867 GWh/yrof national mis-hedge avoided, measured at 3–12 months out. That volume is the part we prove; what it is worth depends on your spread.
At a spread you set, the avoided mis-hedge is worth about£1,933,344/yr
Worked example: a supplier with 5% of GB load at a £10/MWh spread would avoid roughly £1,933,344/yr of mis-hedge (drag the dials to your own numbers). Illustrative — the 3,867 GWh/yr volume is measured; the spread is the example you set; value scales with your share of load.
And that is from public data alone — 18 public feeds, nothing proprietary. With a supplier’s own half-hourly metering on top, the same engine goes further: by region, by segment, by tariff, with the demand signal at its source.
Data & method (the short version)
18 public data feeds, fused. 4,920 days of real demand, generation, settlement and weather across the country — every input public and free, nothing proprietary, everything cached locally.
A machine-learning model. Every accuracy figure on this page is measured against what actually happened, on days the model was not trained on, and against a standard benchmark. The modelling recipe itself is proprietary.
Near-term uses the live weather forecast; long-range uses typical seasonal weather, with wider, measured uncertainty bands — never dressed up as weather-accurate.
National is measured; regional is a clearly-labelled estimate, because no public regional demand history exists to measure against.
GCGridcastGB electricity demand intelligence · a research prototype
Verified on real outcomes18 public data feedsFree public data