One more note - just did my end of year, hourly comparison of my Sense Net (Total Usage - Solar Production) vs. my utility’s billing net meter billing data. As expected, except for maybe 60 or so hours out of the 6220 hours this Sense has been in operation, I have seen hourly results that coincide within +/1% of Sense. The colors show the biggest 25 errors (red), the next 25 biggest errors (yellow), and the next set of 25 (errors):
Removing them, I can generate a linear model with coefficients below, that means SenseNet = 1.000970 x PG&E data - 2.216Wh
If I look at the distribution of the remaining errors they are very balanced around this model, most within 100Wh
And if I look for where the 75 biggest errors occurred in time, that I dropped from my final analysis, most took place during periods where I was having known issues:
None have happened since early October.