Still problems with diffuse radiation

[FlorianK13]

From a rough test @MartGro and me think that the diffuse radiation is still buggy. For example, a south facade has a better yield than a east tilted roof, which should not be the case (At least we think so).

Questions arise:

• Can we use the data / simulator from renewable.ninja
  • https://gsee.readthedocs.io/en/latest/#examples
  • https://www.renewables.ninja/downloads
  • Three-dimensional SOlar RAdiation Model (SORAM) and its application to 3-D urban planning
  • An open-source 3D solar radiation model integrated with a 3D Geographic Information System
    • Do not mention how they handle diffuse irradiance
  • Comparing 2D and 3D Solar Radiation Modeling in Urban Areas
  • Another python package: pysolar
    • Integrating over sky (which we do not want)

[FlorianK13]

We also got a user feedback that the produced electricity of a pv system on an eastern roof top is only 2/3 of the actual production.

[MartGro]

the GSEE python library uses an isotropic sky model, maybe we should also try to use something easier like this.
Then we can take the horizontal diffuse and direct irradiation with scaling based on the time series and know that it should be ok from a scaling perspective (since isotropic is quite easy to verify)

[FlorianK13]

GSEE is a tool to that gives you the following:
For a PV system with given tilt, orientation and area, what is the annual yield. Our idea now is to iterate over tilt in [0, 90°] and orientation in [0, 360°] to get the irradiance for each sky segment [tilt, orientation].

However there is one error in here: For each sky segment we do not get the direct array of light with the proposed method above, but the integrated irradiance (also irradiance that comes with an angle != perpendicular). For our shadow analysis we need only perpendicular light arrays, right?

Or did I understand something wrong here, @MartGro

[MartGro]

So afaik the GSEE uses the Boland–Ridley–Lauret model under the hood. https://lo.unisa.edu.au/pluginfile.php/1161733/mod_resource/content/1/multiple-pred-RENE.pdf
This model gives the proportion of diffuse radiation with respect to the global irradiation.
So overall, the BRL model has no sky model at all I think, because it only gives this one scalar value for the proportion.
I think it would be interesting to check how they are coming up with the effective figures for the tilted panels in the renewables.ninja code, because I guess they have some model sitting on top. The simplest case I guess is assuming that the sky is isotropic, i.e. taking the BRL value at each point in time and than aggregating it (but this would leave us with a completely uniform distribution with no angular dependence, which is maybe not so great).

[FlorianK13]

I talked to a developer of pvlib once and he recommended the Perez Model which comes from this paper. They had implemented this model in pvlib here.

Edit: Just saw that in newer verisons of pvlib this seems to be integrated in this function or one of the other functions from this category.

[MartGro]

Maybe we can take the Perez model for the angular dependence of the diffuse irradiation and the value of the GSEE BRL algorithm for the overall value?

[FlorianK13]

My hope right now is that one of the pvlib functions can be used for both direct and diffuse radiation, so we need only one simulation.

[FlorianK13]

@MartGro and me decided to do the following:

• We use data from https://nsrdb.nrel.gov/data-viewer to get spatially resolved timeseries of Direct Normal Irradiance, Direct Horizontal Irradiance and Global Horizontal Irradiance. This data is measured data, so it includes weather and cloud effects. The Direct Normal Irradiance is measured with a sun tracking device. It is the value [W/m2] that you get directly from the sun in the according 15min interval.
• We sample over random dates from the timeseries and calculate a sky dome.
• This will be the skydome received from the direct part of irradiance.
• To decide the fraction of influence of diffuse and direct irradiance, we compare the timeseries of direct and diffuse horizontal irradiance. We then come to sth like "60% of energy comes from direct, 40% from diffuse".
• For the diffuse part, we keep it simple for now. We will distribute the fraction of diffuse energy equally over the skydome.
• Repeat this for other places to get a grid of sky domes.

[MartGro]

If we want to do a proper time series calculaton, we could also calculate per-month averages for the sky domes and use them for the exact shading analysis for the PV panels.
But this would require explicity simulating them

[FlorianK13]

As commented here:

Ich habe mal meine Anlage 1 mit 16,15 kWp (siehe Signatur) simuliert, für die ich die Planung im SolarEdge-Designer gemacht habe.
Es ist eine O/W Anlage mit viel Ost, etwas West und einem geringen Teil Süd.
In eurer Simulation komme ich in Summe der Teilflächen auf einen Jahresertrag von 7509 kWh.
Der SolarEdge-Designer prognostiziert einen Jahresertrag von 14860 kWh.
Mein bisheriger Ertrag von Mitte März bis heute beträgt 12055 kWh.