Modelling the spectral irradiance distribution in sunny inland locations using an ANN-based methodology. (15th June 2015)

• Record Type: Journal Article
• Authors: Torres-Ramírez, M.; Elizondo, D.; García-Domingo, B.; Nofuentes, G.; Talavera, D.L.
• Is Part Of: Energy. Volume 86(2015)
• Page Start: 323
• Page End: 334
• Publication Date: 2015-06-15
• Subjects: Solar radiation -- Spectral irradiance distribution -- Artificial neural networks modelling -- Average photon energy -- PV technologies; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2015.04.037
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: This work is aimed at verifying that in sunny inland locations artificial intelligence techniques may provide an estimation of the spectral irradiance with adequate accuracy for photovoltaic applications. An ANN (artificial neural network) based method was developed, trained and tested to model the spectral distributions between wavelengths ranging from 350 to 1050 nm. Only commonly available input data such as geographical information regarding location, specific date and time together with horizontal global irradiance and ambient temperature are required. Historical information from a 24-month experimental campaign carried out in Jaén (Spain) provided the necessary data to train and test the ANN tool. A Kohonen self-organized map was used as innovative technique to classify the whole input dataset and build a small and representative training dataset. The shape of the spectral irradiance distribution, the in-plane global irradiance ( G T ) and irradiation ( H T ) and the APE (average photon energy) values obtained through the ANN method were statistically compared to the experimental ones. In terms of shape distribution fitting, the mean relative deformation error stays below 4.81%. The root mean square percentage error is around 6.89% and 0.45% when estimating G T and APE, respectively. Regarding H T, errors lie below 3.18% in all cases. Highlights: ANN-based model to estimate the spectral irradiance distribution in sunny inland locations. MRDE value stay below 4.81% in spectral irradiance distribution shape fitting. RMSPE is about 6.89% for the in-plane global irradiance and 0.45% for the average photon energy. Errors stay below 3.18% for all the months of the year in incident irradiation terms. Improvement of assessment of the impact of the solar spectrum in the performance of a PV module.