The semi-practical method to impose an optimized solar metric into your system (“to integrate the weather”) is:
Isolate a representative panel that you can get to and maintain easily (i.e. clean) and put a dedicated Sense on that one panel (tricky if you don’t have micro inverters per panel). Now that one panel becomes your reference and is essentially tracking the weather … through the limited eye of a PV panel at least.
Stick a PurpleAir or similar near the panel and throw in an anemometer for good measure.
Correlate the dedicated Sense with your hyper-local weather station and extrapolate to the array. Problems should pop right out. My guess is that’s what you really want?
Add a BloomSky for additional data and redundancy. And rainbow watching.
Of course, what you really may desire is forecasting for such wonders as preemptive load-shifting. The more data (dimensions) the better in that case. Like Sense’s disaggregation, it matters less that it works immediately and more that you can grab and archive data for future analysis.
I remind myself of @kevin1’s awesome TOU obsession, which led me to an unofficial challenge to port pricing to a spare Sense CT!. Weather is significantly more multi-dimensional so we’re gonna need a whole bunch of porting for that.