Estimate electrical frequency from clock shift

The Powerwall 2 intentionally runs at between 64Hz and 65Hz when it is supplying power if its battery is full. This has the side-effect of making clocks run faster. It also causes UPSes to run on battery.

Now, the first question you might be asking is how it could be the case that the Powerwall is supplying power and its battery is full. Simple. The house has solar power and the solar power exceeds what the house is drawing. This will mean the Powerwall's battery will charge and, if this condition continues, eventually get full. Since the utility power is off, extra power can't be sold.

Now, you can quickly imagine a problem. The Powerwall's battery is full. The house isn't drawing as much power as the solar panels are creating. Somehow, the Powerwall has got to stop the grid-tie solar system from trying to supply it with power. It does this by bringing the frequency out of specification for the solar panel inverters. Typically, it takes a 64Hz to 65Hz frequency to do this.

So, essentially, this is how the Powerwall shuts the solar panels down when its battery is full and it cannot use all the power the solar system is trying to supply because it cannot sell it to the grid.

Units such as APC and Tripp-Lite UPS detect 60+/-3 Hz as a power good and either reduce AC load outside this frequency or switch off AC charge to rely on battery backup. They also use voltage thresholds.

However, these UPS units have a relatively short backup time compared to the PowerWall2 (PW2). There are UPS's with +/-6Hz tolerance for input power detection such as the Minuteman.

Therefore to reduce non-essential loads that have short-term backup such as UPS, the PW2 runs > 65 HZ intentionally (or > 7.7% fast).

Your clocks were reading (21:00-10:00) * 60min/h = 660 minutes when you were expecting (20:17-10:00) * 60min/h = 617 min or 43 minutes fast = 43/617*100% = 7.0% fast which was the actual increase on the PW2 or 64.2 Hz.

The PW2 was 10% within the 65 Hz that I expected, but I do not have their actual specifications.

I am aware of this but cannot prove it.

45 minutes over a span of 660 minutes is an error of 6.8%. That's rather high. I would have expected a crystal-controlled frequency with an error of 100 ppm or less — i.e., less than 4 seconds over 11 hours.

And power isn't "pushed" by modifying the frequency per se, but rather the relative phase of the various sources.