| dc.description.abstract | Worldwide, the solar energy is on a growing expansion. Although small, if compared to other traditional ways of generating useful energy, there have never been such optimistic indicators about its use and popularization. However, even a self-sustainable energy like this have environmental impact, great economic investment, difficulty in the measurement of all types of solar radiation and problems in the proper maintenance of solar monitoring stations. Knowing that the measurement or estimative of all solar radiation types are crucial for the design and estimative of energy generation in both large concentrating solar power plants or in residential photovoltaic power generation systems, the present work was developed to ensure that, even in continental proportions, the horizontal diffuse fraction can be obtained from the total irradiance with great reliability. The use of Artificial Neural Nertorks (ANNs) with geographical generalization showed better results in relation to empirical models, as well as being more comprehensive than models that use specialized ANNs based on local data. The results showed RMSEr errors ranging from 16,00% to 29,57%, against RMSEr errors from 15,94% to 36,98% in traditional empirical models. The present model also showed a improvement in both local precision and geographic generalization, something that previous models didn’t reach, since when increasing the specialization and precision in a smaller area, there’s a significant loss when using the model in distant and/or larger geographic areas (and vice versa). In this way, the present study ensures that researchers can use ANNs with enough precision and generalization to perform different researches and studies that require the diffuse and/or direct types of solar radiation with greater eco-efficiency and cost-effectiveness in locations that don't measure them or don't have reliable meteorological data. | en |
