Power grids around the world carry a hidden pulse that repeats with remarkable precision every sixty seconds, and it took a team of researchers to trace it back to the most unlikely of sources: the battery backup units that quietly protect hospitals, data centres and office buildings from power cuts. Dirk Witthaut and colleagues report in Nature Energy that a recurring disturbance with a one-minute period appears in grids across five continents, and that it is caused by clock-synchronized self-tests in battery-based uninterruptible power supplies, which inadvertently act in unison. The finding overturns a long-standing habit in power-system modelling, in which small frequency deviations are routinely attributed to stochastic fluctuations in load and generation.
The power system frequency is often described as the heartbeat of an electricity network. In a perfectly balanced alternating-current system, all grid-connected generators and electric waveforms are synchronized, oscillating at a nominal frequency of 50 Hz, or 60 Hz in North America. Yet the power system is never at such an idealized equilibrium. Small deviations from the nominal frequency constantly occur, indicating the imbalance between electricity supply and demand, and they are usually attributed to background fluctuations or noise. The new research shows that at least part of this apparent noise is not random at all, but a coherent signal generated by clock-synchronized actions across continents.
The behaviour of power-system frequency is classically illustrated by the hook curve that follows a generation outage, a pattern that appears in all textbooks on power-system dynamics and controls. A real-world example from Europe in 2004 shows the synchronous frequency falling steeply after an outage, because the slope of the fall, termed the rate of change of frequency, depends on the total inertia of all synchronous generators connected to the system. Frequency containment reserves arrest the decline, which in that event bottomed out at 48.89 Hz, and regulation and redispatch of generation then restore the nominal setpoint. Without such reserves, the frequency would leave its safe operating range, triggering remedial actions such as automatic load shedding. The average frequency is held so accurately that many synchronous electric clocks use it to keep time, and the setpoint is occasionally adjusted, for example to 49.99 Hz in the 2004 record, to correct accumulated clock drift.
Zooming in on such records reveals more than a simple drop and recovery. The frequency curves from Switzerland and Greece do not rise and fall together, and their oscillations vary in timing and frequency, with a distinct East–West continental inter-area oscillation showing up in the measurements. This inter-area mode became particularly visible during the 2006 European grid split. All curves also appear noisy, reflecting fluctuations in load and generation and minor faults. In modelling studies, these small disturbances have been treated as stochastic noise, which captures the background variability that is always present in real grids.
Today’s frequency dynamics are even more volatile than in the 2004 plot, and they are shaped by an increasingly diverse set of physical and cyber disturbances. On the physical side, the growing share of renewable generation is interfaced through power electronics rather than synchronous machines with their large inertia. This gives rise to a range of low-inertia issues: the rate of change of frequency after a disturbance grows steeply, and oscillations emerge from undesired interactions between power electronic converters. On the cyber side, deterministic frequency deviations arise from electricity trading, and data centres, as large and ill-behaved loads, add further oscillations and grid incidents. Meanwhile, the noise in frequency measurements has risen markedly, which has routinely been attributed to lower inertia and increased renewable generation, folded into the stochastic terms of models and rarely interrogated in detail.
The one-minute pattern studied by Witthaut and colleagues cannot readily be explained by random fluctuations alone. Although the phenomenon has long been visible, its root cause remained mysterious until the team identified the culprit: clock-synchronized control actions in battery-based uninterruptible power supplies. These UPS actions generate periodic power-demand peaks, which, due to the wide proliferation of the devices, propagate across multiple voltage levels up to the transmission grid. What appears as random noise in power-system models therefore contains a hidden structure, a coherent signal generated by clock-synchronized actions across continents.
The detective work behind the discovery is particularly compelling. The one-minute signal appears phase-advanced in distribution grids relative to transmission, which localized the source to the low-voltage level. On a research campus, the researchers found that every building showing the pattern housed a battery-based UPS, whereas no building without one did. Their direct measurements then caught narrow current peaks from the battery at the top of each minute, which the UPS manufacturer confirmed were self-tests scheduled by the UPS battery management system. The clincher was a clock timing signature rather than a grid-physics one: with the network time protocol enabled, the phase of the self-tests drifted by a few seconds per day and reset at midnight.
The magnitudes involved are not negligible. The researchers estimate that 80 MW of load switches in unison in continental Europe, causing substantial regulation effort and equipment wear, and at times pushing the frequency to the edge of its safe operating range. The UPS manufacturer, once contacted, issued a software update that removed the strict periodic behaviour, although the UPS units still perform self-tests. The remedy, in other words, was neither expensive nor technically difficult, once the true cause of the disturbance had been identified.
The broader lesson reaches well beyond a single device category. Power systems have always rested on an unstated assumption: millions of small loads behave independently so that their fluctuations average out. Networked timekeeping exposes the limits of this assumption, revealing that what appears unexplained is not necessarily stochastic. Any population of devices that are synchronized by external clocks, from data centres to household appliances, can act as a single large load. Such a load is invisible to planning tools that presume statistical diversity, and when it peaks it can trip protection, induce oscillations, and cascade from there.
Fortunately, the remedy is straightforward and inexpensive: randomize scheduled actions, offset them adaptively, or trigger them based on events rather than fixed timing. Implementing such approaches, however, will require manufacturers of consumer and industrial equipment to recognize that their products increasingly influence how the grid behaves, and to work with grid operators on managing that impact. That conversation, between manufacturers and grid operators about coordinating device behaviour, is worth starting before an even larger synchronized population emerges. The work stands as a fine example of how careful analysis can reveal unexpected cyber–physical couplings that emerge from the interactions between digital devices and power grids.
Subject of Research: A hidden one-minute frequency disturbance in power grids caused by synchronized battery backup self-tests
Article Title: An extra beat every minute
Article References: Dörfler, F. (2026). An extra beat every minute. Nature Energy. https://doi.org/10.1038/s41560-026-02156-w
Image Credits: AI Generated
DOI: 10.1038/s41560-026-02156-w
Keywords: power grids, grid frequency, uninterruptible power supplies, battery self-tests, low-inertia grids, frequency regulation, network time protocol, cyber-physical systems, renewable energy, load synchronization, Nature Energy, grid stability
News Source: Faith Mcneil. (November 6, 2026). Power grids worldwide share a hidden one-minute heartbeat. Scienmag.



