 |
| Titel |
Modelling 1-minute directional observations of the global irradiance. |
| VerfasserIn |
Peter Thejll, Kristian Pagh Nielsen, Elsa Andersen, Simon Furbo |
| Konferenz |
EGU General Assembly 2016
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 18 (2016) |
| Datensatznummer |
250130998
|
| Publikation (Nr.) |
 EGU/EGU2016-11346.pdf |
|
|
|
|
|
| Zusammenfassung |
| Direct and diffuse irradiances from the sky has been collected at
1-minute intervals for about a year from the experimental station at
the Technical University of Denmark for the IEA project "Solar
Resource Assessment and Forecasting". These data were gathered by
pyrheliometers tracking the Sun, as well as with apertured
pyranometers gathering 1/8th and 1/16th of the light from the sky in
45 degree azimuthal ranges pointed around the compass. The data are
gathered in order to develop detailed models of the potentially
available solar energy and its variations at high temporal resolution
in order to gain a more detailed understanding of the solar resource.
This is important for a better understanding of the sub-grid scale
cloud variation that cannot be resolved with climate and weather models.
It is also important for optimizing the operation of active solar energy
systems such as photovoltaic plants and thermal solar collector arrays,
and for passive solar energy and lighting to buildings.
We present regression-based modelling of the observed data, and focus,
here, on the statistical properties of the model fits. Using models
based on the one hand on what is found in the literature and
on physical expectations, and on the other hand on purely statistical
models, we find solutions that can explain up to 90% of the variance in
global radiation. The models leaning on physical insights include terms
for the direct solar radiation, a term for the circum-solar radiation,
a diffuse term and a term for the horizon brightening/darkening. The
purely statistical model is found using data- and formula-validation
approaches picking model expressions from a general catalogue of
possible formulae. The method allows nesting of expressions, and the
results found are dependent on and heavily constrained by the
cross-validation carried out on statistically independent testing and
training data-sets. Slightly better fits -- in terms of variance
explained -- is found using the purely statistical fitting/searching
approach. We describe the methods applied, results found, and discuss
the different potentials of the physics- and statistics-only based
model-searches. |
|
|
|
|
|
|