Relative MBE in 1990 Perez transposition model?

[Mark Campanelli]

Greetings,

  I am looking to quantify the potential bias in the POA-component irradiances as computed from a DNI and GHI measurement, using the Perez 1990 model for the sky-diffuse components. I would like to quantify this as a bounded relative uncertainty interval, such as the nominal computed average component irradiance for a time step +/-4%.

  Consulting the validation study in The Development and Verification of the Perez Diffuse Radiation Model: Technical Report SAND88-7030 (

https://www.osti.gov/biblio/7024029), I don’t find any relative quantification of MBE, so it is does not seem possible to generalize these results to other computations using the Perez 1990 model (unless I am missing something).

  Does anyone happen to know of other validation studies of 1990 Perez model that has more readily generalizable estimates of model bias?

Mark Campanelli

LinkedIn

Intelligent Measurement Systems LLC

Try PVfit today!

[Brighton Mabasa]

Hello Mark,

I did some analysis in South Africa (work is yet to be published)

2 coastal sites and one interior site.

Regards,

Brighton Mabasa

--
You received this message because you are subscribed to the Google Groups "pvlib-python" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pvlib-python...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pvlib-python/CAES-E-75VXCTaz1Rm0fkGvKasUUysQ%3DmwEH%3DFW_cZDYQpRtB8Q%40mail.gmail.com.

[cwh...@sandia.gov]

There is a recent thesis from Stellenbosch University that may help. A paper from that work: https://www.preprints.org/manuscript/202403.0416/v1

The author may have looked at relative MBE.

Cliff

[Mark Campanelli]

Hi Brighton,

  Thanks so much for your response and I apologize for my delayed response.

Your MBE and rMBE appear to be calculated as a measure of the bias of the overall mean values, as opposed to the mean of the collection of biases at all timestamps. Am I interpreting this correctly?

-Mark

[Mark Campanelli]

Hi Cliff,

  Thanks so much for your response and I apologize for my delayed response.

  I think this thesis's analysis is further upstream than I am currently looking, i.e., the decomposition that might be needed before the transposition. What strikes me is that it seems that the uncertainty quoted by vendors like Solcast for remotely inferred DNI, DHI, and GHI can be less that the uncertainty in a non-localized decomposition model based on locally measured GHI data. Do you think I'm interpreting this correctly?

-Mark

[Hansen, Cliff]

I suspect the large MAD values etc. in that paper are more due retaining shaded moments in the data. I don't see how else a reasonable model like DISC can be (on average) 40% incorrect or more.

Sources like SolCast and the NSRDB are pretty good at squeezing down annual average biases. When we looked at an early version of the PSM (the model behind the NSRDB and did hourly comparisons at locations not available to NREL, we found low annual bias but substantial scatter in the hourly errors. Two adjacent pixels had different error distributions. I'm sure the NSRDB has improved with time, but that points out the limitations of statistical validations - you know over time the errors average out, but in any given moment at a selected location, the error can be substantial.

Cliff

---

From: pvlib-...@googlegroups.com <pvlib-...@googlegroups.com> on behalf of Mark Campanelli <mark.ca...@gmail.com>
Sent: Saturday, July 6, 2024 10:30 AM
To: pvlib-python <pvlib-...@googlegroups.com>
Subject: [EXTERNAL] Re: Relative MBE in 1990 Perez transposition model?

 

--

You received this message because you are subscribed to the Google Groups "pvlib-python" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pvlib-python...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/pvlib-python/5cb7a787-78ea-4039-b9d8-78617f96433en%40googlegroups.com.