WARNING: Early June Backfill After Sense June 3 Outage Contains Bogus Data!

[Beachcomber]

You are not showing anything June 22-June 25th where I spotted issues.

Also issue end of day June 14th you missed.

As you eliminated 12AM - 10AM data, that is a huge difference than me as I am bringing the temp way down as I am a hot sleeper.

Based on yesterday (June 25th) where Ecobee and Sense are within 1 minute of the total runtime of ~850 minutes, 41.5% of my runtime was 12AM - 9:55AM ET where your analysis would miss any discrepancies.

[kevin1]

I was just running on the past data. Unfortunately, for me, the new data looks good. Then again, AC hasn’t been running that much and no more back-to-back. I suspect that your issue might be back-to-back hours of runtime. Only have 1 full hour after 10AM for June 6th out here in the West.

[Beachcomber]

Not sure how you would have picked up any discrepancies June 22-June 24th as your AC hardly came on. Clearly you are not in Texas or Florida

[kevin1]

Sorry, I couldn’t do better We’re having chillier than usual weather in Northern CA.

ps: I wouldn’t be surprised if folks in TX and FL were having issues with Sense AC detections due to long back-to-back hours or runtime, though your biggest discrepancies don’t necessarily coincide with longest runtime days. Would be interested in whether which hours on the bad days have back-to-back 60 minute runtimes.

pss: If you want me to run the same analysis I did on your data, PM me the hourly Sense CSV file for June, as well and the Ecobee CSV. Shouldn’t take more than a few minutes.

[Beachcomber]

I don’t think I’ve had a long back to back runtime based on eyeballing the data earlier today.

[Beachcomber]

Unless I am missing an option, it seems the Sense Data downloaded does not supply anything more granular (smaller) than hourly data, nor does it show cycles.

Thus I am not sure how good the info would be when the start time begins one hour and runs into the next

[kevin1]

Thanks to you @Beachcomber, I did a little deeper analysis on my data for the month of June (really for all of 2023, but in the interest of simplicity, kept things to just June.

First off, I charted my Ecobee runtimes, in seconds, for each hour in June, against my Sense AC detected power. My upstairs AC unit uses about 2.5kW, 1 hour (3,600 sec) should show as 2.5kWh and I should get a straight line of point from 0, 0 to 3600, 2.5kWh. And the good news is that I essentially get that, with a little noise, in green and a bunch of outliers, in pink and blue.

The outliers well above the “line”, in pink, are cases where Sense is coming up with an energy number that is too big for a given hour of runtime. Since the Ecobee is likely far more accurate about runtime, I believe that Sense is coming up with a bigger number because it is conflating my higher tonnage downstairs AC (3.5kW) with my upstairs unit (2.5kW).

The outliers well below the “line” , in blue, are likely cases where Sense missed a detection or two.

Here’s a time view on the same comparison, with Sense energy in the orange line and Ecobee runtime in the points with point color using the same legend.

If I look at the numbers, I have conflation 5% of the time, and missed detections about 2.5% of the time. The worst case of a miss happens on June 12th where the Ecobee says my AC was running for about 2 1/2 hours over a 3 hour period, but Sense barely registers any usage. That’s consistent with an issue I saw quite a while back where Sense had difficulty with multi-hour AC runtimes.

                 Number of Hours
 Conflation w          Good          Missed  
Bigger AC Unit       Detection      Detection
      36                619            16

[Beachcomber]

As I have only one unit on a dedicated circuit, I don’t have to worry about confusion between it and another.

However, if you are saying that Sense is confusing detections signature of 2 different devices that use that much energy, that is very troubling imo.

[kevin1]

@Beachcomber, I did a double check and I had blended in a set of downstairs AC numbers against upstairs runtimes, as well as the upstairs Sense numbers. Thought I had joined correctly. I’m the cause of “conflation”, except for one possible instance.

Removing the extraneous downstairs Sense data, the numbers look far better.

 Conflation w          Good          Missed  
Bigger AC Unit       Detections      Detections
      1                653            17

[kevin1]

And a deeper dive on the one point, on June 14th for the 14:00 hour (1PM to 2PM), that still appears to be a “Conflation” sure looks like something else has gone wrong.

Here’s the “golden” Power Meter view of that hour and surroundings.

And here is that same period showing Sense detections of my upstairs AC. It looks entirely feasible.

But the Ecobee data tells a different story. Sense seems to have missed an off somewhere around 2:42PM !

That’s confirmed by a view of the Ecobee graph:

If I try to transcribe Ecobee Ons (green) and Off(red) to that hour, it looks like this.

Two things I notice:

• I have to mentally shift all the Ecobee times to the right a little. There’s a minute or two of delay between when the Ecobee calls for cooling and the board on the furnace triggers the AC compressor. The same when the Ecobee turns the cooling off. I can see it in the the Ecobee to AC / Power meter edge differences.
• Sense has has kept the part marked by the blue arrow as On time for my upstairs AC unit, but looking at my Ecobee data, my upstairs AC went off at the red liner around 2:42PM, almost exactly as my downstairs AC was turning on. I’m assuming that Sense missed the off because of the other unit turning on.

[kevin1]

A couple more device Power Meter screenshots highlight the missed Off transition that caused the discrepancy on June 14th.

Here’s the blower fan for the upstairs furnace
Sense sees it turn off at around 2:47PM

Here’s the AC unit for the upstairs
Sense misses the off at 2:47

And the calculated times based on Ecobee commands

[Beachcomber]

On the brightside, Ecobee Runtime and Sense Runtime on my AC Compressor were the same on 2 days and only off by 1 minute (rounding?) 3 other days.

The rest of the time, well…

[kevin1]

Outside of the two days where we had the outage and backfill (June 2nd-3rd), where it seems like the backfill got the energy right, but still has hosed runtimes and cycle counts, it’s most likely that you are seeing larger scale effects of what I’m seeing below. The green and red bars are where my Ecobee called for AC ons and offs. But my furnace controller clearly inserted delays, between the command and the actual on/off of my compressor. It looks like the off delay runs a little longer than the on delay, but we’re talking about 1 to 5 minutes.

Here are the calculated event times based on Ecobee commands for this hour.

And the furnace fan shows exactly the same kind of timing lag between Ecobee commands and smart plug measurements of my furnace fan. Looks like there is a 3 minute lag between Ecobee off and furnace fan off and a 2 minute lag between Ecobee on and furnace fan on.