PySAM PVWatts returns zero-gen hours after slight perturbation of weather data

Hi there,

I'm using PVWattsv8 within the nrel-pysam python package, version 3.0.2 (also tested on 3.0.1 and same results).

I'm attempting to calibrate a no-financial model against available generation data by iteratively perturbing the irradiance data (DNI & DHI) until converging within an error tolerance. Using this method, I've encountered some odd cases where a very slight perturbation in irradiance data causes some hours to be reported as zeros that were generating previously. I can't determine any differences in the scaled TMY weather data or the hourly Outputs fields that provide explanation for why this is happening.

I put some inputs together to help reproduce -- the two TMY files are the 13th and 14th iteration of the calibration process (it usually converges in 7), so the numbers are very close numerically. The workbook highlights the difference in the "ac" and "gen" fields despite near identical input.

Please let me know if I have neglected to provide any information that could help you look into this issue. Thanks in advance for your time!

Regards,
Ian

Hi Ian,

This might be caused by a bug with the new snow model in PVWatts V8: github.com/NREL/SAM/issues/1137

Please try running the model with en_snowloss = 0. If that does not solve the problem, please attach a sample Python script that generates the results you are seeing from the two weather files.

Best regards,
Paul.

Hi Paul,

Thanks for the quick response. When I turn off the snow loss model, I don't have the issue anymore, which is expected because the solution path to convergence changes.

I re-did my process, printing out the "snow" and "dcsnowderate" fields in addition. I see the same behavior shown in the linked Github issue: dcsnowderate is one of [0, 0.5, 1], and a value of 0 results in no generation. For the hours in question, dcsnowmoderate = 0.5 in one iteration and 0 in the other, so that is indeed the cause!

After reading up on the snow loss model in  this NREL article , my theory is that the slight decrease in irradiance is enough to shift the "snow sliding" inequality (T_a > I_poa / m) from True to False in some hours, which I believe means that snow is still left on the panel, and the output is 0.

With that said, I have two follow-up questions:

1. Now that the issue has been resolved in development, could you provide any estimate on when/how the patched snow loss model will become available to users?
2. Even after the fix is applied, would you expect that this discrete generation jump that I described (depending on whether the snow sliding inequality is satisfied) will still be occur, either to the same or lesser degree?

Thanks again,
Ian

Hi Ian,

The fix to the implementation of the snow model in PVWatts will be part of the next version, which we are working on now and hope to release around mid-November.

With the snow model enabled (and working properly) it is possible that you would see an increase in output from one time step to the next as the snow slides off the array. For example:



And, the same day with the snow model turned off:



Best regards,
Paul.

Thanks, Paul.

I believe I have one last clarifying question. Did the linked issue actually affect the amount of losses due to snow, such that AC output is different before and after the fix, or is it merely the reporting style of the field "dcsnowderate"?

Thanks again,
Ian

Hi Ian,

The issue causes the reported snow loss to be wrong, but the AC output should correctly include the effect of snow.

Best regards,
Paul.