My Power Backup System

I live somewhere the power goes out a lot. Nothing dramatic, just often enough that it stopped being a story and became a scheduling problem.

What I needed

Two things. Switch over fast enough that nothing reboots, and carry the important loads long enough that I am not sitting in the dark waiting for mains.

What I bought

Hybrid, because I wanted the option of panels later without replacing the inverter. LiFePO4 because the bank takes thousands of cycles. The switchover is fast enough that computers don't even blink.

That was the system for three years, and it did the job it was bought for.

What actually happened

The panels went up, and the system stopped being a backup that occasionally runs and became the thing that runs the house.

Two weeks of measured production, first half of August:

Across those fourteen days: 85 kWh consumed, 70 kWh of it from the roof. Solar covered a bit over three quarters of the house.

The pattern is sharper than the average suggests. On days the array clears about 5 kWh, the grid contributes exactly nothing. Not a little, zero. On the overcast run at the start of the month, production halved and the grid picked up the difference. There is no middle: either the sun covers the day or it does not.

The charger is set to Solar only, so the inverter never charges the bank from mains. Utility exists purely to carry load when the battery is spent. That is why the grid column is zero so often. The house simply never needs it.

The thing I did not expect

Adding monitoring changed what I know, not just what I can see.

The inverter talks to a Raspberry Pi over USB; that publishes to MQTT; Home Assistant consumes it. Fifty-nine entities, none of which I have to read manually. I wrote about the plumbing separately in Wiring Home Assistant Into My Solar Inverter.

What it surfaced immediately: I had been through a grid outage without noticing. Not a flicker. An outage that ran long enough for the house to live entirely on solar and battery, while I sat working through it. The inverter had switched over silently, exactly as designed, and no part of the system had any way to tell me. The first thing that noticed was a dashboard I built for a different reason.

That is a strange feeling. The system worked so well it hid its own use from me. It also means I had no idea how much margin I had left, which is a worse problem than the outage.

Runtime, and knowing when to stop

The bank is 5.1 kWh, and the inverter cuts out at 15% state of charge. At the house's overnight draw of roughly 250 W, that is somewhere around thirteen hours from a full battery. Enough to sleep through an outage and be recharging by mid-morning.

Knowing that number turned out to matter for something other than comfort. The homelab runs on this inverter, so a long enough outage does not gently degrade. It drops a running hypervisor cold. I now have a watchdog that reads the battery over MQTT once a minute and shuts the servers down gracefully at 20%, while there is still an hour of margin. Two useful properties fell out of writing it:

• It runs on the hypervisor, not in Home Assistant. Home Assistant is a virtual machine on the box being shut down, and a guard that dies with the thing it guards is not a guard.
• It fails open. If it cannot read the battery, broker down or Pi rebooting, it does nothing. The worst case is the hard cut I already had. Failing the other way would mean powering the house's router off because a Raspberry Pi restarted.

Buying hours instead of only shutting down

The guard is a last resort. It fires at 20%, and everything it does is subtraction. The servers go away and stay away until I intervene.

Between a full bank and that point there is a lot of runtime to be bought, and none of it requires turning off anything I would miss. That became obvious during an outage that ran three days, with the battery working down into the low forties while I watched a dashboard tell me how long I had left.

So the house now sheds, in an order that matches what I would actually give up:

45%   dispenser        nobody notices
35%   TV               switched at the socket, so standby goes too
30%   living room system
25%   deep freezer     last, because food
────────────────────────────────────────
20%   the guard shuts the servers down
15%   the inverter cuts everything

Only during an outage. On mains a deep cycle is just a deep cycle, and nothing should be switching itself off because the sun was poor.

The freezer sits deliberately at the bottom. It is the largest continuous load and the most tempting thing to cut first, and it is also the one whose cost is measured in spoiled food rather than inconvenience. Everything cheap goes first.

A clock-driven layer sits under it: the freezer rides its thermal mass between 03:00 and 07:00, and the dispenser is off overnight. The two layers collided in a way worth recording. A reconciler re-asserts the schedule every fifteen minutes, because a missed switch-on is invisible; you find out by opening the freezer days later. But during shedding that is exactly wrong: it would switch the freezer back on four minutes after a battery crisis switched it off, and neither automation would look broken on its own. So shedding raises a counter, and the reconciler stands down while it is above zero.

Three years ago this was a box that kept the lights on during cuts. It is now the primary supply, with the grid as the fallback. That is the inverse of what I designed, and I only found it out by measuring it.