Tiago Brandão, portfolio

Electricity use in Espinho, Portugal, from November 2020 to April 2026

The winter that broke the forecast

Monthly electricity consumption in Espinho from November 2020 to April 2026. Consumption peaks every winter; in winter 2025/26 it rose higher, and January 2026 was 10.5 percent above January 2025.
January 2025
+10.5%
January 2026 against January 2025

I built a simple forecast of Espinho's monthly electricity use from open data. Over 30 months of testing it missed by 2.5% on average. Then, in the winter of 2025/26, consumption ran up to 10.5% above the year before and the forecast fell behind. This page is how I tried to find out why.

Tiago Brandão

Computer Engineering student at ISEC

01

Test the forecast

Move across the chart to compare each forecast with what really happened. Every month is forecast using only the months before it, as it would be in real life. Switch methods to see how the error changes.

Month
January 2026
Actual
9.80 GWh
Forecast
8.91 GWh
Error
9.1%

Actual consumption Forecast November 2025 to February 2026
2.48%

Mean error over the 30 test months

1.62%

Mean error in the 26 normal months

8.05%

Mean error, November 2025 to February 2026

I kept the 2-year average. It beats last year's value in 15 of the 26 normal months: a real but small edge, and it could be luck.

02

Three suspects

An error of 8% in four months in a row is not noise. I checked the obvious explanations one at a time.

The cold

Partly. Between 11% and 43% of the increase, depending on how it is measured.

I measured cold with heating degree days, using Eurostat's definition on daily temperatures for Espinho. January really was colder. December and February were as cold as the year before, and consumption still rose 10.3% and 8.0%.

Across Portugal's 278 mainland municipalities, the ones that got colder jumped more: the single-rate jump was +12.1% in the quarter with the least extra cold and +16.5% in the quarter with the most. A fit across municipalities puts the weather at about 43% of the typical jump (95% range 36% to 50%). That compares municipalities with each other, so it is not proof of cause.

Heating degree days in winter 2024/25 and 2025/26, with the change in consumption

November
+8.9%
December
+10.3%
January
+10.5%
February
+8.0%
2024/25: 51, 228, 206, 1772025/26: 143, 227, 242, 180

Something only in Espinho

No. The whole country shows it.

E-Redes splits consumption by tariff. The jump sits almost entirely in simple-tariff contracts: 80% of Espinho's increase in November and 89% in December. Time-of-use contracts barely moved, and Portugal as a whole shows the same pattern.

December 2025 against December 2024, by type of contract

Simple, Espinho
+17.4%
Simple, Portugal
+15.2%
Time-of-use, Espinho
+2.4%
Time-of-use, Portugal
+1.1%

November has the same shape: simple +14.8% in Espinho and +14.3% in Portugal; time-of-use +3.5% and +1.0%.

A billing glitch

Unlikely to be the whole story.

E-Redes data is billed consumption, so a billing error could fake a jump. But REN, which runs the national grid and measures electricity on the network rather than on bills, also recorded a rise. Espinho still sits 2 to 4 points above the country.

Change against the same month of the year before

November
+8.9%
December
+10.3%
January
+10.5%
February
+8.0%
Espinho, billed (right)Portugal, national grid: +6.6%, +6.9%, +7.6%, +3.7%

What's left: for the typical single-rate jump of +14.4% in November and December, about 6 points are weather and about 3 look like ordinary growth, which leaves about 5 points, roughly a third, unexplained. More customers and the smart-meter roll-out don't explain it. It concentrates in simple-tariff contracts, in Espinho and across the country. With monthly open data I can't tell whether that is real consumption in that group, an effect of estimated readings on bills, or something else. I'd rather say that than invent an answer.

03

How much was the cold?

From November to February, Espinho used 3.06 GWh more than the winter before. Two honest ways to measure the cold in Espinho give two different answers, so I report both. A third way, comparing Portugal's municipalities, gives a higher share: about 43% of the single-rate jump in November and December.

0 GWh3.06 GWh, the whole increase
0.33 GWh

Explained by the cold, 11% of the increase

2.73 GWh

Left unexplained

2.5 MWh

More consumption per extra degree day

A model fitted only on data from before November 2025, comparing each month with the same month a year earlier.

04

What didn't work

  • Correcting for recent growth

    3.41% error against 2.69% for last year's value, over a 42-month test. It reacted too late.

  • Monthly mean temperature

    Its link with consumption is mostly the seasons. Year on year, one degree colder adds only about 0.8%.

  • Heating degree days

    Best overall at 2.29%, no better in normal months (1.62%), and only with the real temperature, which a real forecast wouldn't have.

  • Hourly data

    The open hourly data for Espinho stops in September 2023, so an hourly forecast isn't possible yet.

05

Limits

  • Monthly, billed data

    Not real-time consumption. May 2026 was left out because it was incomplete.

  • A short test

    30 test months, four of them an unusual winter. The differences between methods are small.

  • Rough comparisons

    E-Redes measures billed consumption on the distribution grid; REN measures what the public grid supplies.

  • Everything checked

    Each number here was confirmed against the data or its source. What I couldn't verify, I say so.