Feat/inverter availability

[kandersolar]

Code for inverter availability calculations. Two main functions:

• is_online, to detect when inverters are offline
• downtime_losses, to estimate lost production

Detecting offline inverters should be easy, except communication outages make it tricky. The code assumes that meter-level data is reliable and never has comms outages but is mostly robust to inverter comms outages. I believe the only thing it will fail at (assuming good meter data) is when some inverters are truly offline and some are only non-communicative. In that case the code says that they're all offline.

For lost production, the preferred method is inter-inverter comparison -- examining how inverters have performed relative to each other in the past and assuming the same relative performance can be used to calculate lost production when some inverters are offline. If all inverters are determined to be offline, an expected_power signal provided by the user is used to determine lost energy. If there is no expected power data for the outage (eg the entire system is offline, so there's no weather data), a user-provided 12x24 typical production profile is used instead.

Documentation links:

• https://rdtools.readthedocs.io/en/feat-inverter_availability/availability_example.html
• https://rdtools.readthedocs.io/en/feat-inverter_availability/generated/rdtools.losses.html#module-rdtools.losses

---

Edit for anyone reading this -- the above info is mostly about the original implementation and is largely out of date.

• Code changes are covered by tests
• New functions added to __init__.py
• API.rst is up to date, along with other sphinx docs pages
• Example notebooks are rerun and differences in results scrutinized
• Updated changelog (waiting for Feat/monthly soiling metrics #193 to get merged with the whatsnew file)

[kandersolar]

Picking this PR up again to implement the methods in Anderson & Blumenthal "Overcoming communications outages in inverter downtime analysis", PVSC 47

[kandersolar]

why they get rotated on the left plots but not the right

It's the difference between series.plot(ax=ax) and ax.plot(series). Pandas has a default set of matplotlib formatting params it uses when you do series.plot(ax=ax). The function now uses pandas plotting everywhere so that the formatting is applied consistently: https://github.com/NREL/rdtools/blob/feat/inverter_availability/docs/system_availability_example.ipynb

[mdeceglie]

@kanderso-nrel I suggest adding the reference for your work on the availability algorithm on the overview docs page

[kandersolar]

@kanderso-nrel I suggest adding the reference for your work on the availability algorithm on the overview docs page

Done (despite brain fart of "soiling" in the commit message). I also took the liberty of adding one for the degradation section.

[mdeceglie]

I think this is looking good. Here are a couple suggestions.

[mdeceglie]

LGTM Thanks for putting this together @kanderso-nrel !

[mdeceglie]

Should we use double backticks in the tables describing dataframes?

[kandersolar]

@mdeceglie I added an "other uses" section in the notebook with an example like we discussed in the meeting earlier today, of running the entire thing at the inverter level instead of at the system level: https://rdtools--131.org.readthedocs.build/en/131/system_availability_example.html#Other-use-cases

The other option (omitting the cumulative energy data) would require some more in-depth changes to get working, so I think we should leave it as a possible future improvement.

Also made a small change to how the outage bounds are reported; the end of the outage was previously being reported as the timestamp of the first non-outage value rather than the last outage value. This was introducing a small error into the calculated actual energy during the outage, because it was including that single interval of energy after the outage.