| Summary: | Facing the energy transition, solar energy, whether thermal or electric, is currently one of the most viable alternatives, due to its technological maturity and its ease of operation and maintenance compared to other renewable energies. However, before its implementation, it is necessary to assess its potential. Remote sensing represents one of the low-cost solutions for solar energy assessment. Nevertheless, cloud cover is a main problem when validating the data. This study identifies satellite GHI profiles that cannot be used in energy production simulation. The validation is performed using parametric and non-parametric statistical tests. From the profile identified as invalid for simulation purposes, a site-adaptation methodology is proposed based on statistical learning using the machine learning algorithms “Best subset selection” and “Forward Stepwise Selection”. Linear and non-linear heuristic models are also proposed. The final AS7 model is selected through RMSE, MBE and adjusted R<sup>2</sup> indicators and is valid for any sky condition. The results show an increase in R<sup>2</sup> from 0.607 to 0.876.
|
|---|