polito.it
Politecnico di Torino (logo)

Solar radiation forecasting with deep learning techniques integrating geostationary satellite images

Raimondo Gallo

Solar radiation forecasting with deep learning techniques integrating geostationary satellite images.

Rel. Edoardo Patti, Alessandro Aliberti, Marco Castangia. Politecnico di Torino, Corso di laurea magistrale in Ict For Smart Societies (Ict Per La Società Del Futuro), 2021

Abstract:

The energy produced by photovoltaic (PV) systems is deeply tied to the amount of received solar radiation. The intermittent nature of the solar power represents the primary problem for an efficient integration of PV systems into the electricity grid. The prediction of future solar radiation allows to quantify the energy production of the PV system, hence guaranteeing a more reliable and stable energy supply. Nevertheless, the availability of remote sensing imagery and machine learning techniques opens new possibilities to improve the solar radiation forecasting. In the literature, several studies have been done integrating satellite images to sophisticated deep learning models with the intent of improving the prediction of solar radiation forecasting, however the majority used satellite images to estimate cloud presence, cloud motion or to generate a solar radiation maps integrating ground measurements. In this work, it is presented a novel approach to estimate solar radiation, for a given location, directly from multi-channel satellite images (provided by the SEVIRI instrument of the METEOSAT geostationary satellites) coupled with ground local solar global horizontal irradiance (GHI) measurements. The nowcasting and forecasting of the solar radiation are accomplished through the implementation of deep learning techniques. More in detail, a 2D convolutional neural network (2D-CNN) is used for the nowcasting, while a 3D convolutional neural network (3D-CNN) is employed to perform forecasting. The obtained results show an improvement in the forecasting performance for prediction horizons greater than 30 minutes and up to 5 hours, compared with techniques using historical solar radiation measurements. On the other hand, the nowcasting results are also promising but they didn't achieve the awaited precision. Nonetheless, the results proved that utilizing deep learning techniques with multi-channel satellite images as direct input for solar radiation forecasting is a valuable and promising alternative to more complex models; moreover, the proposed solution can be easely extended to new study areas since satellite imagery, the primary input to the proposed models, potentially cover most of Earth's surface and they are extremely easy to retrieve.

Relators: Edoardo Patti, Alessandro Aliberti, Marco Castangia
Academic year: 2021/22
Publication type: Electronic
Number of Pages: 73
Additional Information: Tesi secretata. Fulltext non presente
Subjects:
Corso di laurea: Corso di laurea magistrale in Ict For Smart Societies (Ict Per La Società Del Futuro)
Classe di laurea: New organization > Master science > LM-27 - TELECOMMUNICATIONS ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/21225
Modify record (reserved for operators) Modify record (reserved for operators)