Battery Self-Tests Cause a One-Minute Dip in Grid Frequency, and Jülich Measured It on a 1.5 MW Unit

Grid frequency in Europe dips every 60 seconds, and a team at Forschungszentrum Jülich says the likeliest cause is battery systems that check their own state of charge at the start of each minute. The group published the finding on October 30, 2026, in a press release that cites a Nature Energy paper. The release says the dips are measurable and unusually regular but not critical to grid security.

The Terminal could not open the Nature Energy page, which sat behind a cookie redirect. The measurements below come from the team's arXiv preprint (arXiv:2510.09862, submitted October 10, 2025), whose authors include Dirk Witthaut, head of the Networks and Complex Systems department at Jülich's Institute for Energy Systems Engineering, and Arndt Neubauer of the transmission operator TenneT TSO GmbH. Where the published version differs from the preprint, this piece will be out of date.

What the 1.5 MW unit does every minute

The cleanest evidence is one machine. Jülich's campus holds an uninterruptible power supply backed by a battery storage system rated at 1.5 MW and 0.5 MWh, with three battery racks in parallel on the DC link. The preprint reports "narrow current peaks recurring every 1 min" on the DC side, and says the manufacturer confirmed they come from the unit's periodic self-tests.

The number that matters here is the denominator. A 1.5 MW pulse is trivial against a continental grid; the question is how many identical pulses land on the same second. The preprint names no manufacturer, but says the units in buildings 1 and 4 came from different makers than the one in building 3, so the pattern is not one vendor's quirk.

The team instrumented the campus with more than 80 power-quality meters, each with 0.1% accuracy at the fundamental frequency. Meter 023, on the battery transformer, showed a sharp one-minute peak in both active and reactive power in both measurement windows (March 24 to July 3, 2024, and October 1 to November 5, 2024). Four more meters showed a one-minute peak in active power.

Meter group (Jülich campus)	One-minute peak
023, battery transformer	Active and reactive power, both windows
006, 007, 051, 088	Active power
004, 005	Small, first window only
014, 015	Broad peak near 70 s

Why a clock makes it worse

Self-tests that fire at the same second across thousands of sites add up. With network time protocol (NTP) synchronisation enabled, which the campus switched on after October 8, 2024, the preprint finds the daily waveform drifts by roughly 5 to 7 seconds per day. Synchronised clocks are what turn many small loads into one coherent one.

That is also why the authors expect the problem to grow. They write that the periodic variation in power imbalance should increase as large battery systems join the grid and smart household devices with synchronised clocks spread.

How big is the dip, and where

The preprint's amplitude figures are regional and thin. In the Nordic grid the amplitude "regularly exceeds 10 mHz"; against a 50 Hz nominal frequency that is 0.02%. Weekly averaged amplitudes rose by a factor of 1.3 to 1.4 when the COVID-19 lockdown began in April 2020. The authors report larger increases in the Great Britain and Nordic grids and a mild one in Continental Europe, using series that end in 2020 or 2021.

The press release adds a claim the preprint does not contain. It says that in Ireland, data centres with large battery systems account for about one-fifth of national electricity demand, and that the amplitude of the one-minute signal quadrupled between 2016 and 2021. That is a single-source figure, and it ties a demand share to a signal amplitude without showing the link.

1. 20161 x
2. 20214 x

Source: Forschungszentrum Jülich press release, October 30, 2026, accessed 2026-11-01

What it costs and what fixes it

The release says the brief demand spikes make hydroelectric turbines slow momentarily and then speed up again, which adds wear. It gives no cost figure, and neither does the preprint.

No binding standard sets how often a battery system should test itself. Witthaut suspects manufacturers picked 60 seconds as a sensible interval and the top of the minute as the test time. The proposed fix is a software change: offset the test, either by an amount that changes daily or by a random one. The preprint contrasts manufacturers on this point, saying some keep a perfectly periodic schedule.

The release also names air conditioning, heat pumps, washing machines and dishwashers as possible contributors, per Witthaut, who says they "may also exhibit similar patterns." The preprint's own tests are less generous: a smart oven showed a one-minute pattern only when run on a timer, and a laptop showed no one-minute pattern.

What would change this read

The causal chain has one measured link and several inferred ones. The measured link is a 1.5 MW unit whose maker confirmed one-minute self-tests. The inferred links are that other sites do the same, that they are aligned closely enough to move national frequency, and that the effect is global. The release says it "appears to be" the cause, which is the right strength.

Three observations would shift the argument. A grid operator that publishes frequency data before and after a major battery operator adds a random offset would test causation directly. Manufacturer confirmation across more than one vendor would settle how widespread the schedule is. A measured wear figure for hydro turbines would turn a nuisance into a cost.

Until one of those arrives, the finding is a well-measured campus case with a plausible, unproven extrapolation. For the load side of the same grid, see our reporting on US battery storage capacity and duration and on four-hour batteries competing with gas peakers across 43 markets.