Identify how changes to settings affect performance

This is somewhat similar and somewhat different to what I’ve written before, so sorry for any redundancy. I’d love some way to reasonably estimate how changes to my thermostat settings impact runtime, based on my own real data, with some way to remove the effect of other variables like outdoor temp differences day to day.

An example of what I’d hope to answer is the age old debate about setbacks, and whether it saves money (or at least runtime) to have a certain setback overnight, while at work, etc. Intuitively, it seems like there should be a way to get a reasonably solid answer on an individual basis, as it’s probably a different answer for each home and each hvac system.

I’d ideally like to run a group of settings for a week, a month, whatever, and then change them and let them run for a second week, month, etc. and have Beestat be able to tell me that one or the other was X% more efficient, correcting for outdoor temperature variations.

Basically, it seems to me that there are a huge number of settings and variables that will all produce adequate and subjectively indistinguishable comfort, but they can’t all be equally efficient.

Maybe this is too complex to be realistic, but then again y’all are the data wizards and I’m not. So just throwing it out there.

1 Like

I like the idea, but let me expose a few problems:

  1. There are too many variables. Ideally you want to do this in a short time span to keep the outdoor temperature consistent, but even then it’s so variable with cloudcover, rain, wind, etc that it would be impossible to do much with the data. If you averaged over longer periods to account for that then you just get different outdoor temperatures.

  2. It’s a lot of work to actually plan and do all this analysis. Even with beestat helping it’s still going to be a lot of manual tweaking.

The amount of people that would use a feature like this would be so small (probably just you :slight_smile:) that it’s hard to justify.

However, I still like the idea. To make it worthwhile we have to account for the above. I think the solution could be AI. You could theoretically train an AI that has inputs for different weather factors, system types, home types, settings, etc and then either allow the user to change the inputs to look for optimizations or else just randomly change things and attempt to brute-force some recommendations.

Would it work? Maybe. It’s certainly worth a look and I would love to see the results. Will I do it? No clue…I’ve always found AI interesting and would love a project as a catalyst to learn more.

Hahaha I figured that might be the answer but then again I’m impressed by what you CAN do already so who knows.

My line of thinking was, if Beestat already displays an average temp for the day, and you only compare days with the same average temps, eventually (not sure how large a sample would be required), those intra-day variations would weed themselves out. For example, if through the month of January you run your system with a setback, and then through February you run at a constant temp. At the end of those two months maybe there were 6 days in January with an average temp of 34 and 5 days in February with an average temp of 34… Average out the runtimes of only those with a direct comparison and see if a meaningful difference emerges.

But I’m prepared to believe it’s a lot tougher than that. Thanks!

And I guess it wouldn’t be that hard for me to just do that in a spreadsheet, if it actually worked anyway.

If you do it let me know. You could use the beestat API easily enough to scrape some of the data if you wanted to pull it down into a spreadsheet. Off the top of my head things I would also want to consider are: outdoor temperature, sunlight, precipitation, daylight hours, wind, humidity, indoor requested temperature, sensor placement, sensor participation, external heat sources (space heaters, ovens, etc), open windows/doors.

I’ve always been interested in studies like this, especially ones to find out optimal heat pump settings. Just never had the time or the will to collect all the data.

I’m currently air sealing and insulating my rim joist in my house, and have found areas of extreme cold air movement. So I know the air sealing and insulating will help.
What would be nice is a feature that could compare the pre vs post change slope graphs. This could be used for HVAC replacement or other insulation & air sealing projects.


My graph shows a little chaotic a month post change. Last week was actually showing a negative heat slope. Each weekly update clears out a bit of old data and replaces with new data. Looking at the standard temperature slope we see actual data (higher than 35 degrees, pre-change) is lower than the theoretical slope, so I know my change is very effective almost acting like new equipment or even a new house!