Making Sense of 'Always On' - Chapter 2 - 'Always On' Blips

I had a few odd blips in Always On the past few days. The first one was just before the Smart Plug support was announced.

Not sure what to make of it. Anyone else see this sort of thing?

Source is hourly downloaded data.

Can’t help you, my friend. My Always On has been pretty steady over the last few weeks, though there have been a few points of interest:

  • Always On has dropped steadily in neat increments throughout the smart plug rollout as I have added new smartplugs to my system and Sense starts accounting for the new “known loads”

  • Always On ticked up starting this weekend after we installed several HEPA air purifiers due to air quality issues from the Camp Fire.

  • A few drops due to data dropouts that are also visible in the Power Meter. Most of the dropouts are probably related to late night updates of Sense firmware. I’m going to poke a little bit more into the longer data gaps at the Nov 15th/16th boundary.

I’m on my mobile device so I don’t have access to my always on but I’ve been having issues with always on reading high. I put my server and refrigerator onto smart plugs thinking it would drop my always on consumption but instead my always on has stayed the same and I have additional bubbles for my smart plugs, so these devices are being counted twice. There are many times thoughout the day where you can add the bubbles up and in total they use more than my actual power draw. I’ve been trying to find a fix but haven’t been able to remedy this.

Strange. I thought they said they would take any consumption from smart plugs without underlying recognized devices away from Always On. Maybe it’s just not enough to be noticeable over 24 hours.

I have a bad feeling about Always On, regardless. It doesn’t look like it is a measurement of anything. It’s just derived from other measures. I guess the idea is useful in trying to highlight devices that are left on. But it doesn’t seem to be rigorous. I can imagine that the minimum total usage over 24 hours might sometimes/often reflect always on use. But you can imagine cases that would confuse it. And the calculation was more complex than that, though it hasn’t been disclosed. And now the smart plug consumption comes out of it. I guess that is to try to keep device totals from exceeding the total draw, but it seems more obfuscating to me.

Other got a similar treatment. It is the difference between device and Always On totals and the total draw, but zapped to zero if it would otherwise be negative.

To me, the only ‘real’ data is the device & plug consumptions and the total draw. There may be times when the device consumptions are off due to missed on or off events. I don’t think there is anything we can do about those other than just being aware and knowing that the phd’s are doing their best.

I find it interesting because I don’t think I have actually seen my Always On ever go screwy on the bubble display, but clearly there are some anomalies when viewed via the downloaded version of the data. Looking at my data, it looks like spikes, both downward and upward in the downloaded Always On are visible artifacts of underlying Sense “real data” problems. Virtually all my downward spikes are indicators of data dropouts. Upward spikes show up in places where I saw “negative total usage” due to my Sense probe flipping out.

I think what happens is that for the downloaded hourly value, Sense averages the total Always On accumulated for the hour, and in the process highlights missing data and erroneous data. Not to say that your situation is the same.

Here’s my history - note all the spikes. Many do coincide with data dropouts and other errors.

Below is a more detailed view of my Always On history between Jan 18 and mid-Aug 18, annotated with all the largest Sense vs. utility errors I found back in this thread.

  • 58 hours with significant data gaps
  • 36 hours with negative total usage
  • 125 hours with significant differences between Sense and my utility for other reasons

There is definitely some correlation here between Always On hourly irregularities and Sense / utility discrepancies.

I’ve added 7 smart plugs and my Always On hasn’t changed much.

but have you added them to device that are “always on”? Adding a smart plug to your washer that was already recognized won’t help :slight_smile:

My 4 new 110s should be here Friday and I’m planning on adding them to my home theater power strip, my home office power strip, electronics closet power strip (cable modem, Hue Hub, SmartThings hub, RPi, etc, etc) as all of those have “always on” devices on them. Not sure what to use my 4th one for yet. I was going to use it on my Dishwasher but since it just got recognized I no longer have that need really.

Yes, the 2 devices that I moved onto the smart plugs are “always on”.

All of them but one was added to things that are always on such as, modem, router, switch, server, RPi, desktop computer, and AV stuff.

I added one to my LG OLED TV that Sense had found. The numbers are now accurate for the TV.

OK, one more pointer for all us data analysis guys. Don’t forget to add the blips that you might never see. I was scratching my head because I wasn’t seeing all the error data points I expected when I matched up Always On with my previous error analysis. I was only seeing 5 (out of 36) Negative Total Usage hours and 55 (out of 94) Data Gap hours. It turns out that Sense export or the calculations behind hourly Always On dropped 99 hours from Jan 1st through today. When I add those hours in with 0 kWh for Always On, an even more compelling picture emerges.

Look like one can locate/predict 95% of the data gaps and negative total usage hours from quick analysis of the Always On export data !

I also have noticed that after moving some of my “always on” loads to smartplugs, the sum of all bubbles almost always exceeds the actual realtime consumption. I would have expected the always on value to decrease to account for the load shifted into now-known bubbles due to smartplugs. In my case, the network rack that I’ve put on a smartplug accounts for over 50% of my always on usage, but the always on bubble has not decreased by a corresponding amount.

@RyanAtSense can you weigh in on this discrepancy that many of us are having with smart plugs and always on?

@pswired, @senseinaz,

I don’t think that Always On automatically goes down when you add smartplugs. It’s all about what the low-water mark looks like for a 24hr period. Here’s a graph of my exported Always On vs. all my smartplugs. With most of my smartplug additions, there was a small reduction in Always On within 24hrs. But when I added my smartplug for my Hot Water Recirc Pump that draws 45W, nada ! Why ? Because it’s on timer, and is off during the quietest part of the night.

FYI -The Toto toilet and the Upstairs Furnace are smartplugs that were merged with existing devices, hence the earlier history.

Always on should automatically go down if you take those always on devices and move them to smart plugs. The low water mark should get lower by whatever wattage the plug accounts for. Both of my devices that were moved to smart plugs have a constant power draw and are never turned off.

I agree with @senseinaz that always on should be reduced by the amount of the smartplug usage that has been deemed “always on” in the evaluation period. I am in a similar boat where much of my smartplug load is always on. Currently, this causes double-reporting in the bubbles, and artificially deflates the value of the “other” bubble.

I agree, if indeed your Smartplug device(s) truly has a constant baseline, 24/7. My point was that not everything on smartplugs has a constant draw, so timeframe is also a part of the calculation.

So, my always on value decreased last night by a small amount. Not enough to entirely account for the new smartplug-tracked always on loads, but some. Here is an example of a new smartplug-tracked load that, in my mind, should have cut the always on value by ~150w:

I’ll get back to everyone on this. Lots of folks are out early for the holiday. Let me speak with engineering and data science and I’ll return to this thread next week.

Here is a snapshot that may help:

Looking through my real-time history for the last few days, my baseline always-on usage (not the bubble-reported amount, but my minimum usage manually collected from the real-time graph) is 360w.

I am out of town and can easily observe the conditions when the house is at minimum usage. Here are screenshots during such a period:

The only loads active during those screenshots are the always-on bubble and the smartplug-tracked loads. The actual always-on usage is 360w, as reported by the power meter. The sum of the bubble values is 435w.

Something is being double-counted.

Remember that Always On is a number based on a 24hr interval so there could be situations where your Always On from the previous 24 hour period is higher than your current “always on” (Always On doesn’t update immediately upon seeing a lower always on value). Your historic Always On, as well as all your identified usage, gets subtracted from your measured total usage to determine Other. In that situation, Other will be negative, and total usage will be less than the bubbles show. It might be helpful to look at the exported hourly data for Always On to see what’s happening.