CEC 6-parameter coefficient generation

[cdeline]

SAM is able to extract the CEC parameters required for calcparams_desoto. This is done through the 'CEC Performance Model with User Entered Specifications' module model, and coefficients are automatically extracted given nameplate parameters Voc, Isc, Imp, Vmp and TempCoeff. The method is based on Aron Dobos' "An Improved Coefficient Calculator for the California Energy Commission 6 Parameter Photovoltaic Module Model ", 2012

Ideally we should be able to work with the SAM open source code, extract the bit that does the coefficient generation, and put it into a PVLib function that would allow users to run calcparams_desoto with any arbitrary module type. At the moment we are dependent on PV modules loaded into the SAM or CEC database.

Thank you!

[janinefreeman]

SAM solution routine is located at https://github.com/NREL/ssc/blob/develop/shared/6par_solve.h , function "solve_with_sanity_and_heuristics" . Additional dependencies on other files 6par_newton.h, 6par_search.h, 6par_jacobian.h, 6par_lu.h, 6par_gamma.h, all located in https://github.com/NREL/ssc/tree/develop/shared .

[mikofski]

I'd like to try to take this up sometime soon, maybe by the end of the year? Is anyone else working on it? I haven't seen a PR. Does anyone have any strong opinions? Is seems like there's a SAM implementation, is there a PVLIB-MATLAB version already? thx

[cwhanse]

I'm not working on it. If it is practical, it would be great to wrap the SAM method. NREL put a lot of effort into the heuristics for initial values and updating. We have a function in PVLib for MATLAB but it's a very different algorithm. For consistency with the parameter databases it would be better to use the SAM algorithm.

[adriesse]

Do I read between the lines that the two algorithms do not converge on the same parameters? Then it would be of great interest to have python implementations of each! Or perhaps even more important: documents describing the full algorithms along with every heuristic, assumption and constraint they implement...

[cwhanse]

@adriesse yes. Each model fitting algorithm I'm aware of, will produce different parameter values from the same data although the predicted IV curves may be similar.

[markcampanelli]

FWIW I am trying to establish a common/fair metric to compare different model fits in PVfit. In addition to choosing, say, which residual(s) to analyze, the effect of model discrepancy significantly affects such comparisons. PVfit’s orthogonal distance regression (ODR) fits to the single-diode model (SDM) often are worse than other algorithms’ fits in terms of terminal current residuals, but I have observed consistently that this is likely due to model discrepancy. One might claim that ODR is being more “honest” here, but I think it really means that the model should be improved. Note that PVfit’s ODR alternatively optimizes the residual of the sum of currents at the high-voltage diode node in the equivalent-circuit model, and it doesn’t make unfounded assumptions about error-free measurements in all data channels except the terminal current (which also has implications for parameter non-identifiability). I have seen plenty of evidence that ODR fits are “good” in the absence of significant model discrepancy, such as for some (but not all!) double-diode model (DDM) fits across various material systems.

[wholmgren]

If it is practical, it would be great to wrap the SAM method.

Two kinds of practicalities here:

1. Technical. I don't know how to do it but I am sure it's possible. I am ok with it so long as pvlib remains straightforward to install from source.
2. Legal. We would need assurance from NREL that it's ok for pvlib to distribute this SAM code under the terms of a MIT license rather than SAM's mixed MIT/GPL license.

[janinefreeman]

In the next 6-8 months, the NREL team will be updating our python wrapper for the SAM software development kit so that you can pip install the sam-sdk and call its routines in a much more native pythonic fashion. That might be a quick way to implement the CEC parameter generation to solve both the technical and legal practicalities Will mentions above, as well as the many file dependencies involved with that routine. The SDK is licensed separately from the SAM open source code under an MIT type license, so that distribution with pvlib would be ok, and once we have it set up as a python package, that should make it a non-issue to include with pvlib installation from source.

To @adriesse 's point, the publication associated with the method implemented in SAM is here: http://solarenergyengineering.asmedigitalcollection.asme.org/article.aspx?articleid=1458865

[wholmgren]

Thanks @janinefreeman that is great news!

[markcampanelli]

@janinefreeman Does NREL's web site no longer provide free preprints of Aron's paper?

[janinefreeman]

@thunderfish24 I'm not finding a copy of it in NREL's pubs database, but it looks like you may be able to get it here (didn't create an account to give it a try): https://www.osti.gov/biblio/1043759

[adriesse]

@thunderfish24 I sent you a preprint.