- 81.1 TWhelectricity used in 2024values × their length in hours
- 324.5 TWhthe same year, values added up4.0 × the real figure
- +4.35%error on a 23-hour dayfor “average × 24 hours”
- 356quarter-hours missing in 2024about 89 hours, none in 2022–23
Four times too much
Fingrid publishes Finland's electricity consumption as an average power for each interval, in megawatt-hours per hour. Until 13 June 2023 each value covered one hour; since then each covers 15 minutes (Fingrid, dataset 124 (opens in a new tab)). Adding the values gives the energy only while every interval is one hour long.
One hour at an average of 9,000 megawatt-hours per hour is 9,000 megawatt-hours of energy. Published as one hourly value, adding the values gives 9,000. Published as four quarter-hour values of 9,000 each, adding them gives 36,000: four times the energy. Multiplying each value by its length in hours, a quarter, gives 9,000 again.
Energy per year: 2022 79.220 terawatt-hours, 2023 79.409 terawatt-hours, 2024 81.127 terawatt-hours. The plain sum of the published values: 2022 79.220, 2023 208.129, 2024 324.508. The two agree in 2022, when every value was hourly, and part company from 13 June 2023, when the values became quarter-hourly.
On every one of the 528 hourly days the sum of values equals the energy; on every one of the 567 quarter-hour days it is exactly four times the energy. Between 2023 and 2024 the energy used rose 2.2%. The added-up values rose 55.9%. A dashboard built the obvious way would show Finland's consumption jumping by half in a year in which it barely moved.
The switch was not even simultaneous inside Fingrid: Finland's imbalance settlement moved to 15 minutes on 22 May 2023 (Fingrid, 17 March 2021 (opens in a new tab)), three weeks before this series did. A report that joins two such series cannot assume one time step.
Which day?
The series is published with UTC timestamps. A Finnish day starts at midnight in Helsinki, two or three hours before midnight UTC, so a report that groups by the UTC date moves the first hours of every day into the day before.
| Over 1,096 days | Median | 90th percentile | Largest |
|---|---|---|---|
| Difference, MWh | 445 | 1,410.2 | 10,032 |
| Difference, % of the day | 0.21% | 0.62% | 4.48% (2023-03-26) |
Over whole months the two largely cancel out: the largest monthly difference is 0.15%. For daily reporting, peak days and anything compared day by day, the day boundary has to be the local one.
When the clocks change
Twice a year a Finnish day is not 24 hours long. The last Sunday of March has 23 hours and the last Sunday of October 25. A shortcut that multiplies the average by 24 is exact on every other complete day (1,016 of them, error 0%) and wrong by exactly one hour's share on these.
On the six days the clocks change, the shortcut average times 24 hours is off by: 2022-03-27, a 23-hour day, 4.35 percent; 2022-10-30, a 25-hour day, -4.00 percent; 2023-03-26, a 23-hour day, 4.35 percent; 2023-10-29, a 25-hour day, -4.00 percent; 2024-03-31, a 23-hour day, 4.35 percent; 2024-10-27, a 25-hour day, 3.23 percent (incomplete day).
The last of the six, 2024-10-27, is also missing 7 of its 100 quarter-hours, so two errors overlap on it.
Missing quarter-hours
1,021 of 1,096 Finnish days have every value they should. No day is missing entirely and no interval appears twice. The gaps are all in 2024, most of them in October and November.
Missing quarter-hour values in 2024 by month: Jan 0, Feb 0, Mar 0, Apr 0, May 1, Jun 9, Jul 3, Aug 4, Sept 36, Oct 116, Nov 178, Dec 9. 356 in total. No values are missing in 2022 or 2023.
Two days carry overlapping values, which a plain sum counts twice:
- 13 June 2023, the switchover: the last hourly value (12:00–13:00 UTC) overlaps the first quarter-hours, which start at 12:30, by 30 minutes.
- 26 June 2024: 97 quarter-hour values instead of 96; the extra one sits off the 15-minute grid and overlaps its neighbours by 14 minutes in total.
The yearly total for 2024 is therefore understated by the energy of about 89 hours, and the 2.2% increase is computed on incomplete data. The page reports the gaps; it does not fill or estimate them.
How the pipeline gets it right
Each fix sits in one layer, and each is tested on 23- and 25-hour days, the UTC and local split, both resolutions, gaps, overlaps and extra values before it runs on the real data.
Fetch: one request per month to the Fingrid API, with the key in a header and the publisher's rate limit respected; a changed download fails instead of silently replacing figures. Silver: one row per interval with its UTC start and end, its length read from the row itself, and its energy computed as value times hours; intervals that are neither 15 nor 60 minutes, and negative values, go to quarantine. Gold: totals by Finnish day using the official time-zone rules, the length of each day, and a completeness flag per day with gaps and overlaps listed. Report: reads energy and the local date, never adds up the raw values, and shows completeness next to every total.
- FetchOne month per requestKey in a header, rate limit respected, a changed download fails loudly.
- landing/
checksum per fetchchanged data fails - landing/
- SilverOne row per intervalUTC start and end, length from the row, energy = value × hours.
- grid_
interval - quarantine
15 or 60 min only→ quarantine - grid_
- GoldTotals by Finnish dayDay length from the time-zone rules, completeness and overlaps per day.
- grid_
daily_ totals
observed = expectedflagged incomplete - grid_
- ReportEnergy, not valuesTotals read energy and local dates, with the completeness flag beside them.
- daily
- monthly
- yearly
- daily
What this doesn't prove
- The figures are the values Fingrid published on 14 October 2026. One download cannot show whether the 2024 gaps are later filled or past values corrected.
- The series is a grid balance (production plus imports minus exports, with a small estimated share), not metered consumption. Nothing here describes any customer, sector or company.
- The data shows where values are missing, not why.
- Nothing here explains why consumption changed between years; the yearly totals exist to show the size of a counting error.
Four questions for your own reporting
- Does every interval series you report on carry its own start and end, or is its length assumed?
- Are your daily totals grouped by the local day, or by the date in the timestamp?
- What do your reports show on the last Sundays of March and October?
- When a value is missing, does the report say so, or does the total just come out lower?
If any answer made you pause, start a conversation.
Data and sources
Source Fingrid / data.fingrid.fi, license CC 4.0 BY (opens in a new tab). Electricity consumption in Finland, dataset 124 (opens in a new tab), fetched through the Fingrid Open Data API on 14 October 2026. Changes: daily, monthly and yearly totals, completeness counts and the error figures above are derived by Giggabytes from the published values; the values themselves are unchanged.