Energy Usage History

A few improvements in my house convinced me that I needed to update my SimplePlot script to work on a weekly basis, rather than daily…

  • Our Model S started being detected again (Model 3 still MIA since June)
  • New AC units installed in July seem to be detected regularly now
  • Data dropouts have been cleaned up

Now I just need to figure out what to take aim at: The top 10 devices, plus Other, Other Identified and Always On.

SimplePlot2.R (3.2 KB)


I’ve been pondering how to make an always-Always On argument for fridges/freezers and an interesting plot is putting the watts on a log scale. I wonder if you order the stack bottom-to-top in lowest-to-highest usage whether it makes anything pop?

1 Like

Very interesting. How many days does this cover ? It looks like you sized and “alpha’ed” the “points” so that they would all be visible even with overlap. Gives you a view of the broad dynamic range of some devices and the regularity of others. The log scale spreads things out nicely by order of magnitude.

Yes, 50% transparency for the points.
The sizing is a little arbitrary but was going for clarity in making the datasets discrete/pop.

62 days ==> 1488 points max

Washer/Dryer (Wemo): 1452
AC (Wemo): 833
PoE switch (Kasa): 1488
Fridge (Wemo): 1486
OLED TV (Kasa): 312

The data needs some filtering. All devices are on smartplugs but the difference in the plugs comes through. The true “offs” need to be clear for visualizing Always On. There are also some outliers that could be purged.

My main takeaway though was a little unexpected and wrong! … though I can’t stop thinking there’s something to it … that energy data follows some sort of meta-Benford’s law. Putting things on a log chart of course immediately gives that false impression! And in this case the data kind of lined up in a way that exaggerates that impression. Data danger!

1 Like

Here’s the same with the the bottom 3 fixed as Always On, Other and Other Identified. The rest are ordered by the coefficient of deviation (sd/mean).

SimplePlot2.R (2.9 KB)

1 Like

OK - a few weeks later and something interesting is emerging…

  • The ordering of the major detections is different because the coefficient of variation for each has changed for the window period, so the color key is all different. But weekly totals are all the same EXCEPT…
  • The Electric Vehicle detection, which I renamed Tesla Model S, has back-propagated into my data. This data was not there before, so the data science team at Sense has been at work doing detections after the fact (Yay!)


Updated Energy Usage History
After not-just-a-few late nights of experimenting with ways to to get more accurate detection of my EV charging, I can now update my weekly energy history charts with accuracy EV data, one by kWh and the other by %. You’ll notice that my Sense data had some disruptions in July 2019 and Sept 2019, but all is well now. Only had to make one minor change to the SimplePlot2 R program.

I track the 10 largest usage devices, where 2 of the devices are really rollups of remaining devices:

  • Other Identified - is the total energy for all the other detected devices not in the top 8
  • Other - is all the unidentified energy

The “wasteful” Dryer in olive, is really our dryer heating element plus two of the largest in-floor heaters we added back in Oct 19. All are conflated in to a single Dryer device. There are a couple more less-used floor heating elements lurking in the Other Identified.

The “Tesla Charging” in blue, represents the data supplemented from my home-grown detector. I could probably separate out by Model S, Model 3 and Both, but why bother ?

Two nice things have happened over the past year. I have been able to get “Other” down to a much smaller %, and it looks like I have also been able to reduce by YoY June usage. BTW - That dip in late June into July '19 was vacation. Not going anywhere this go round.