Explorations in Home Automation
January 10, 2024
One of my motivations for doing home automation has been to keep the pipes from freezing at a remote property. In winter we reduce the heat at 63F to be economical (not wasteful) but high enough that we don’t get mold everywhere (which seems to happen below about 60F). The crawlspace is pretty well sealed and unless the outside temperature dips below perhaps -10F for a long period of time it will stay above freezing if the house is reasonably warm (70-75F). I should probably put a space heater with thermostat in the crawlspace, but for now I’m using waste heat from the house to keep the crawlspace above freezing, and this is only needed a few days a year.
Several years ago I installed a Particle Electron, which is an Arduino-compatible, cellular connected (3G, very inexpensive service) device. I outfitted this with a DS18B20 temperature sensors, drilled a small hole in the floor (under the fridge), and dangled the temperature sensor in the crawlspace. Cellular coverage is not great at this location, so I upgraded the Taoglas antenna that came with the Electron with a slightly larger one. Still the connectivity is spotty, but I have this report (using MQTT) to my Home Assistant (HA) server every 5 minutes though I have seen some long (>1h) gaps so may need to revisit this. So far, at least when the temperatures are low and it matters, this has not had an impact. The Electron is plugged into a wall outlet, but also has a LiPo battery, so I check where it’s getting it’s power as well, and if it’s drawing from the battery that tells me the power is out at the property, and I won’t have Internet thus no way to adjust the temperature (this also triggers a MQTT message with a topic associated with a boolean on HA, so I track the power and get automated alerts if it goes out. Fortunately that is an edge case that has only materialized once or twice in 25 years so far.
In addition to reporting crawl space temperature, the Electron will send a warning message (using the MQTT topic mapped in HA to a boolean) when the crawl space drops to 37F, and a similar warning message when the temperature hits 35F. I had automations to send me texts, turn on lights, etc. when these were received so that I could manually increae the heat via remote access to my Nest thermostat.
While we were experiencing the deep freeze around here in January 2024 I decided to automate these adjustments. (Having said this, of course it makes much more sense to put a heater with a thermostat in the crawlspace! But for now , partly because we almost never reach any danger levels, I’m using waste heat from the house as an interim hack.)
The code is github.com/cecat/Crawlspacer/ and also has some internal logic to send warning messages if the temperature hits 35F and again if it hits 32F. (more on this below)
The Electron gives me crawlspace temperature, and for outdoor temperature I use a HA weather plug in (Meteorologisk Institutt), though there are several like this. In this case, based on the Lat/Lon you configure it finds the closest National Weather Service station.
I implemented a rules-based control system using a finite state machine approach with four states, each a ‘danger level,’ zero through 3. State changes are each triggered by outdoor temperature (OT) and crawlspace temperature (CT) thresholds. When moving into a given state the thermostat setting is increased or decreased:
| Danger Level | Outside Temperature | Crawlspace Temperature | Set HVAC to: |
|---|---|---|---|
| Level-0 | T >= 20F | n/a | 63F |
| Level-1 | 10F < T < 20F | n/a | 68F |
| Level-2 | T <= 10F | > 39F | 70F |
| Level-3 | T <= 10F | <= 39F | 73F |
| ALERT 1 | n/a | <=35F | |
| ALERT 2 | n/a | <=32F |
I kept the original alert messages for 37F and 35F, which are shown as the extra alert levels at the bottom of the table. As before, these trigger more agressive automations than just SMS messages, including turning on a warning light in our bedroom, since these indicate that human intervention may be needed. It’s a 2 hour drive (if I’m home!) so if we hit these temperature levels I decide whether to go drain the pipes. If I decide correctly then I never get the warning message that the temperature dropped to 32F!
I also explored (along with a buddy named GPT-4) other methods such as ML models or PID, but even with 2+ years of data the subset that is useful (when the temperatures are very low) is quite tiny from a model training point of view. I did do some analysis, for the periods where the outdoor temperature was below 40F, of the correlations between outdoor temperature and crawlspace temperature (pulling the latter down) as well as between the crawlspace temperature and indoor temperature (pulling it up). In very round numbers, the crawlspace will drop 1 degree for every 4 degree drop outside. And the crawlspace temperature will increase one degree for every 3 degree increase in indoor temperature.