I disagree that these data are “wild and crazy”. In Fourier space, they look quite reasonable:
Where I’ve processed a day of data, ensemble averaging using 10 minute windows. We were not home, so there was no human-generated noise.
The spectrum shows a broad band of low-frequency hum (5 to 20 s periods), which you can plainly see in the time series posted above. The energy decays with a slope of -2.25, which is too much for there to be any persistence (i.e., no fractal properties). My estimate of the noise floor of the instrument is 1e4 below the peak indicating we are getting good precision from the instrument.
I have no idea what the low-frequency hum is. I think it’s too low for the human ear to detect…