A Tale of Two Houses

A couple months ago another Sense user asked for help to see if he could figure out the the relationship between outside temperature and energy usage during his Florida peak cooling season His ultimate end goal was to figure out whether one Ecobee thermostat schedule was more energy-efficient than a second one. This forced me to revisit earlier experiments here:

House #1 - Florida

Our Florida user’s data initially looked like this, with five pieces of data per day - date, cooling energy (kWh), average outdoor temp, average relative humidity and schedule (left column data vs right column data). He was using color to look for correlation between different levels of energy usage vs the other three factors.

Initially, I suggested that he do two things:

  1. Use the CDD (cooling degree day) methodology for finding the relationship
  2. Organize the data in a date / data column format where the thermostat schedule (A or B) is just another column of data.

Here’s the same data after that reorganization. You’ll notice that is is still sorted by thermostat schedule first, and date second.

Once the data is organized like this it’s pretty easy to figure out the CDD for each day - C65 means you just subtract 65 degrees from the mean outdoor temperature. Once that’s don, it’s pretty easy to chart kWh vs CDD for both schedules (A/B). The cyan and orange are the fit lines for each schedule and the gray shaded shapes are statistical envelopes (95th percent confidence level) around each fit line.

Two things come out of this chart.

1.There’s a pretty good correlated fit. Energy usage looks pretty linearly correlated with CDD. R^2 for fit is about 0.88 where a perfect linear fit is 1.0.
2. There’s not a lot of difference I energy usage vs CDD between the two schedules. The slopes and intercepts are close the same.

A month later, our Florida user had taken this analysis to the next level by collecting more data, including inside temperature and using the daily outside / inside average temperature difference (Diff = Outside Temp vs. Inside Temp), instead of CDD, as the x variable in the relationship. That, for him showed up even better correlation.

And it should give good correlation if the primary route for heat into the house is conduction - outside temperature causing heat flow into the cooler inside. Physics (Fourier’s law) says that the heat flow into the house is proportional to the difference between the outside temp and the inside temp (dT/dx) as well as the exposed surface area of the house (A). And presumably the energy needed to pump that heat energy back outside again, is somewhat proportional to the heat coming for the temp inside to remain fairly stable.

Going back to the data, morphed to the long data format, it looked like this.

And charted, it also showed a linear relationship between kWh and Diff, with an R^2 of 0.89, just slightly better than vs CDD. And again, very little difference between the Ecobee cooling schedules.

House #2 - Northern California

This new comparison inspired me to go back to my Ecobee and Sense data, and update them, to do the same kind of analysis. In my case, I have 4 years of data, all at different time intervals, albeit with some missing data. Sense’s does have a nice daily export fo this work, but Ecobee’s main output is in 5 minute increments. Also, my Sense didn’t start really nailing my AC usage until after the 2019 cooling season. For that reason, I decided to rely entirely on my Ecobee data, including cooling runtimes (as a proxy for kWh) to do the analysis.

My first try was a mess, but I intended it that way. 4 1/2 years of daily Ecobee data for ALL the seasons, starting from April 2018. Notice that I have two AC compressor units, Down and Up, for each floor in my house. Too much data, wouldn’t you say. One other note - I’m converting cooling runtime into kWh knowing exactly how much my single-stage compressors use per second thanks to Sense.

One other thing that immediately stood out to me was that whereas the Florida users was only cooling when the daily average outside temperature was higher than the inside temperature, my house / Ecobees were cooling even when the average outside temperature was up to 15 degrees cooler than the inside average. And no, I don’t have a heat pump - that is all cooling. More discussion on this later. Let’s clean up the data first.

Cleanup step 1 = Remove broken AC data. I replaced my downstairs AC unit in the July 2019, but it have been failing for a little over a month. I corralled the data for the time my AC was known to be bad, and segregated it. Pretty clear that the failure cost me lots of excess energy usage (and this is based on runtime - the malfunctioning compressor might have been eating even more energy per second than normal operation). But that raises and additional question - did I miss any bad data ?

With a little more cleanup and reorganization I can get a better idea how much of 2019 was a cooling “bust” when it came to energy usage. Here, I have done two things:

  • Pulled out just the summer cooling data - May 30th - Sept 15th
  • Separated out the “AC Failure Down” data from “Down” and “Up” and plotted by year.

Pretty clear that most of 2019 (Green) was a problem, at least up until the AC unit was replaced, because the slope of the Down for 2019 is still steeper and starts higher than all the other years, just like the known bad period… I’m also thinking that the failure of the Down unit, forced the Up unit to work harder and run longer since 2019 is the most energy intensive cooling season for the Up as well.

So next, let’s try removing all of 2019, including the AC failure to see the data without potential outliers.

I’m also seeing another hidden trend in this data. I see a bunch of data points on the zero line for 2018 that seem to bring down energy usage that summer. Turns out COVID and camp is to blame. COVID for keeping us around the house from 2020-2022, and our daughter’s 3 week summer camp for keeping us traveling away from home during the summer of 2018. The Ecobees have motion-driven away feature. Now that I have data that seems to pass the smell test, I need to explain why my house needs are conditioning even when the daily average temp exceeds the inside temp on a regular basis - Next Posting.

1 Like

Thank you for this.

1 Like

The other half is still coming - comparison with a Northern California house.

Part 2 - Cooling When the Outside is Cooler Than The Inside ?
My home data exposed a real challenge for me - why are we running the cooling system when the average daily outside temperature is cooler than the inside temperature ? A few theories popped into my head:

  • Is the daily average inside and outside temperature really a good measurement for my house ? Our Bay Area summers have warm days and cool nights, so the outside temperature typically oscillates between above and below the inside temperature.

  • How much of the inside heating is pure conductive vs. via solar radiation ? Both my attic and outside stucco hit temperatures far above the outside ambient air temperature due to solar heating. Is there a way to separate out the solar contribution ?

Here’s a complicated start to answering to the first theory. I charted the Difference between outside temperature and inside temperature throughout every summer day based on 5 minute samples. The gray to black shows the range of paths through the day with the black highlighting the most traveled Diffs. The blue shows the average Diff for each 5 minute sample and red shows the median for the Down and Up thermostats.

I can see that on average, there is only a short stretch of the day, from a little after noon until about 5PM where the outside air temperature is higher than the inside temperature, though some some days that period can be far longer and some summer days the outside stays cooler than the inside.

Since I can see a fairly stable pattern, though with a lot of variation, I’m going to chart out a comparison between a couple of the mean values for every 5 minute sample through the full history of summer days. I’m going to look at the mean temperature difference vs. the mean power usage for each 5 minute period.

I can quickly see two main differences.

  • It looks like the cooling power curve is shifted about 2 hours later than temperature difference curve. In otherwards, changes in the temperature difference take about 2 hours to pass through to affect the amount of power needed to cool.
  • There’s another factor affecting power usage - THE THERMOSTAT SETTING ! You can see the spot at 5-6PM where my schedule lets the inside temp drift up a degree or two. Plus two mysterious spikes and just before 3 and 4PM in the Down usage. What gives ?

I took a quick look at my Down schedule in Ecobee-land and noticed a few things. The baby blue line is the set temperature and that is based on two things - the schedule on top (different black/blue/orange time ranges with different temp settings), and runtime adjustments. Runtime adjustments can be anything from manual changes that stay until the next scheduled change, to automated eco-friendly adjustments made by the thermostat, in green. The two spikes come from the Ecobee trying to do a little extra cooling downstairs before we move into the next more expensive ToU period (up until 3PM is off-peak, up until 4PM is partial off-peak). So those two down turns in the set point just before 3PM and PM are automatic runtime adjustments for lower cost.

I have also experienced the need to continue cooling my house when outside temperature is 5 to 10 degrees lower than my inside temperature. Part of that is thermal mass of the house, which you allude to. You also mention sun shining on the house and in through the windows. I believe another contributor is energy consumed inside the house: electricity used inside the house ends up as waste heat, plus our bodies generate heat as we metabolize food.

1 Like

@jefflayman, here’s a little useful happenstance. I decided to look at my Beestats (3rd party better UI for Ecobee) and I spotted some interesting things in the vein you suggested. I know that Ecobee temp sensors are not extremely accurate and tend to underestimate temperatures swings as you get away from their sweet spot (68 to 72 degrees), I spotted a weird reading - my Living Room, on the Down thermostat, was peaking near 100 degrees (probably 105 considering Ecobee range compression). Turns out that was because the sensor was in the window sun during a daytime window. Shows the influence of the sun.

And as you would expect, my Up thermostat also shows the longer radiant heating period of my attic, to probably 105 degrees given the Ecobee sensor range compression, again.

Wow ! A partial answer to the question that got this whole discussion with the Florida user started - what’s the most energy-efficient schedule to achieve my cooling goals ? The answer is that “it depends” based on the physics of your AC system and the heat capacity and insulation of the house.

1 Like

And then Sense follows with this article that explains why my Ecobees appears to have a life of their own.

1 Like

They state humidity is a factor.

Of course my 3 months of data was wrong as the humidity sensor was very far off.

“The answer boils down to how energy intensive it is to remove heat from your home. It’s influenced by many factors, such as how well your house is insulated, the size and type of your air conditioner, and outdoor temperature and humidity.”

1 Like

Definitely a factor as well. Someday we might get to a point where our smart thermostats run a set of experiments on our house (hopefully while we are gone) and come up with a heating and cooling model for your house / HVAC combo that you (or it) can use to create the most efficient schedule to meet your constraints. Beestats for Ecobee already does a very simple model, but doesn’t consider most factors yet.

@jefflayman, I also discovered one more reason that my house was cooling when apparently the inside temp was warmer than outside - the Ecobee measure for outside temperature is reading higher than my exact location. even though my town is small, we can have a 10 degree differential between the area closer to the SF Bay vs. the hills. On hot days like today, the difference is especially pronounced - I was seeing a 69 read from the most local weather station (in Palo Alto, not Menlo Park) when the Ecobee, and Apple weather were both quoting 74.