Researchers Trace 60-Second Power Grid Fluctuations to Battery Software

Researchers in Germany have identified a possible explanation for a strange pattern that has been appearing in electricity grids around the world. Every 60 seconds, the frequency of the grid changes slightly, with the same disturbance repeating at almost exactly the same point each minute.

The fluctuations have been observed for years, including by German transmission system operator TenneT. They are generally too small to threaten the electricity supply, but their regular timing has puzzled engineers trying to understand where they come from.

Now, researchers from Forschungszentrum Jülich, the University of Cologne and Karlsruhe Institute of Technology have traced at least part of the problem to software running inside battery systems. Their research, published in Nature Energy, shows how regular self-tests performed by uninterruptible power supplies (UPSs) can create short increases in electricity demand that eventually reach the wider grid.

The interesting part is that the equipment isn't necessarily doing anything wrong. The batteries are simply checking their operating condition, as they have been programmed to do. The problem comes when large numbers of devices perform those checks at the same time.

A Battery Check That Reaches the Power Grid

The researchers started by examining frequency measurements from electricity grids across five continents. They found the same narrow 60-second pattern in 12 different grids, including those serving continental Europe, Britain, Australia and several smaller island networks.

Electricity grids normally operate at a nominal frequency of either 50 Hz or 60 Hz, depending on the country. That frequency changes slightly as electricity demand rises and falls. When consumption exceeds generation, the frequency tends to decrease, before the grid's control systems bring it back towards its normal value.

In this case, the changes were extremely small, often measured in millihertz. What caught the researchers' attention was their timing. Ordinary fluctuations in electricity demand are rarely so precise, yet this pattern continued repeating at exactly one-minute intervals over several years.

That suggested something was operating according to a clock.

The team eventually found a useful clue at Forschungszentrum Jülich's own research campus, where they examined electrical measurements from buildings and equipment connected to the site's distribution network.

Four buildings showed a particularly clear 60-second pattern in their electricity consumption. All four had battery-based UPS installations, while the buildings measured without these systems showed no comparable signal.

A UPS normally provides backup power to equipment that cannot tolerate an unexpected interruption, including servers, computers and industrial systems. Its battery management electronics also carry out routine checks to monitor the condition of the batteries.

When the researchers examined one installation more closely, they discovered short current peaks occurring at the battery terminals every minute. The equipment manufacturer subsequently confirmed that the peaks were caused by scheduled self-tests within the battery management software.

The team then followed the electrical disturbance through the UPS and the building's supply system. Although the effect wasn't always obvious in the raw AC current measurements, analysis of the recorded signals revealed the same one-minute pattern at successive points in the electrical network.

It could eventually be traced through several transformers to the campus connection with the public 110 kV grid.

That's quite remarkable when you consider where the disturbance started. A small change in current inside a battery system was still detectable after passing through the electrical infrastructure supplying an entire research campus.

The researchers also investigated the timing of the battery checks. Some UPS installations produced a power-demand peak shortly after the beginning of each minute, while another showed a pattern that drifted over time.

By changing whether the equipment's clock was synchronised using the Network Time Protocol, the team observed corresponding changes in the timing of the electrical disturbances.

It appears that battery manufacturers have independently adopted similar schedules for their diagnostic routines. There is no universal requirement for these checks to happen at the beginning of every minute, but it is a convenient interval for software developers to use.

Individually, these brief power changes are unlikely to matter very much. Across thousands of installations, however, they can begin to add up.

What Happens When Thousands of Devices Follow the Same Clock?

The scale of the effect becomes more interesting when looking at the wider electricity network.

According to the researchers, a frequency variation of just 3 millihertz in the continental European grid corresponds to a power imbalance of approximately 80 megawatts. This is an estimate of the combined effect across the network, rather than the consumption of any individual battery installation.

And while the frequency variations are relatively small, they are not completely without consequences.

Hydroelectric power stations are among the facilities affected by these fluctuations. Their turbines and control systems have to respond whenever the balance between electricity generation and demand changes. A disturbance every 60 seconds might be tiny, but it still gives the regulating equipment something to correct, and those repeated adjustments can add to mechanical wear over time.

The researchers also found that the one-minute pattern has become more noticeable in certain parts of the world. Ireland is a particularly interesting example, with the recurring frequency drop increasing from around 3 millihertz in 2016 to 12 millihertz in 2021.

During those years, Ireland was also experiencing rapid growth in data centres, which typically use large UPS installations to protect their servers against power interruptions. By 2022, data centres were responsible for approximately 18% of the country's electricity consumption.

The study doesn't prove that Ireland's data centres caused the entire increase, although their growing electricity demand and extensive use of battery backup systems make them a possible contributor.

There's another consideration here, too. The electricity grid isn't quite the same as it was when most power came from large rotating generators. Those machines have a considerable amount of energy stored in their spinning rotors, which helps cushion sudden changes in electricity demand. The frequency can still fall, but the stored energy slows the rate at which it changes.

Solar generation and many battery installations connect through electronic inverters, which do not inherently provide the same mechanical inertia.

As the mix of generation changes, a given disturbance in electricity demand can therefore produce a larger frequency response, depending on how the network and its control systems are configured.

Interestingly, the researchers don't believe UPS installations are necessarily the only devices responsible for these patterns.

The study also identifies periodic power consumption associated with other digitally controlled equipment, including a smart oven and clusters of Raspberry Pi computers. Electrical appliances, heat pumps and other connected devices could potentially produce similar effects if their internal routines are synchronised.

This is where the problem becomes particularly relevant to electronics engineers.

A software developer might decide to check a battery's status every minute, while another programmer schedules a sensor reading or power adjustment at the same interval. Neither decision is particularly unusual, and the resulting change in electricity consumption might be insignificant for a single device.

But if thousands of products use the same clock and perform their routines at the same moment, the combined electrical load becomes much harder to ignore.

Fortunately, the solution may be relatively straightforward.

Rather than running diagnostic routines at a fixed point each minute, manufacturers could introduce small random delays or vary the timing between checks. The equipment would still perform the same monitoring functions, but the resulting changes in electricity demand would be spread across time.

The researchers have already worked with one UPS manufacturer on a software update that removed the strict periodicity of current peaks in part of the system. The improvement was incomplete, with some self-tests still occurring at discrete times, but it demonstrated that changing the software could reduce the effect.

Grid operators have also made progress by improving their frequency-control systems. Changes introduced in Scandinavia in 2023, for example, have helped reduce the amplitude of the recurring fluctuations, although those measures address the grid's response rather than the timing of the devices causing them.

I think this is a particularly interesting example of how embedded software can have consequences far beyond the equipment it was written to control.

There is nothing inherently unusual about checking a battery every 60 seconds, and the individual installations examined by the researchers were operating as their manufacturers intended. Yet the combined effect of these perfectly ordinary routines was large enough to be measured across national electricity networks.

As more batteries, inverters, EV chargers and connected appliances become part of the electricity system, engineers may need to pay considerably more attention to when their products consume power, rather than simply how much.

Sometimes the difference between an electrical disturbance and completely normal operation could be as simple as choosing a different moment to run a piece of software.