Using my Vera Lite and Z-wave home energy monitor I was able to record a week’s worth of electricity consumption at 30 second intervals. I was surprised how hard it was to use the data to do anything, but amazed with how my behavior changed when I knew our household consumption was being tracked. Again, if you want to improve something, I feel it needs a tracking system. Electricity usage is a nice case study for this, because it has a clean metric (wattage) over a time series.
Unfortunately, the raw dataMine plugin for Vera couldn’t deal with things like outliers and didn’t have a good ability to zoom and I had to resort to MATLAB to pull the data together. See this gist for all my source code.
After some conditioning, you can put everything on a plot like this:
The plot does show some cyclical usage data and taught me that the shape of each day is determined by the laundry usage. I was also very surprised how much leaving on computers at night changed the floor for each day, with some nights showing that we continually used about 800 watts, where other nights we were down to 100 watts while everyone slept. This was due to us leaving on kids lights as they slept in addition to keeping electronics powered on. Also interesting, that roughly 10% of the time, we were using more than a standard dev over this. It looks like the washer and dryer are huge contributions.
How do we compare?
One interesting thing this analysis forced me to do was to compare my usage to the national average. On average, US households use 30 kWH a day. While the variance on that must be huge and our seven-member household is probably at least double the national average, inspection of our electrical bills shows that we have averaged 40.6154 KWH a day or 1692 Watts every hour. On one hand, this is $5 a day with our local rates, or roughly $1,800 a year.
For greater insight, I used my favorite data visualization language (d3.js) to pull together the following interactive visualization. It is pretty fun to use and allowed me to zoom in to a day and slide that across the window to see usage habits. Be sure to check it out and select the bottom to zoom in on my electricity usage.
You can also click on the picture below to look at each leg separately and see what interesting patterns you can spot.
However, I had to do some more work to see any actual patterns. First, I built the plot below after writing some code to pull out each day. For each day, I show the raw data with a red line superimposed that is the result of a smoothing filter (robust local regression using weighted linear least squares and a 2nd degree polynomial model, with a coefficient of 0.1). If you follow the red line below, you see the bump every morning for breakfast as the heat turns up and lights turn on. Otherwise, the clothes dryer drives a lot of the electricity usage, as does another spike for dinner. These plots allowed be to better understand our home as a system that is most active for breakfast and dinner. Our house works hardest to produce a load of laundry, not good for a family with four young kids and two very active parents. What else do you see in the plots below?
Another helpful plot, lets me look at the total usage per each day. It is pretty surprising to me to see how much our daily usage varies. It will be nice to look at this with more data and see if there is a good trend there. (Or decide that we don’t care and at least keep the trend from blowing up.)
The hardest chart of all to produce was to put the smoothed averages all on one x-axis. Here you can see the common rhythms for each of our days.
Enjoy, and please share your thoughts in the comments. After taking time to make the plots, I’m out of time to do the analysis.
If I had more time, you can eventually get to something like this:
Cool Stuff for later work
- You can see the ranges that appliances use at energy.gov
- Life could be easier if we could get this into green button format and use Bigely
- time series handout
- We also might be able to see plotWatt
- We can potentially upload per these instructions
- We need to learn this
- If we watched this, we could better understand time series analysis matlab webinar
- Wikipedia’s page on Nonintrusive load monitoring