Servicios Personalizados
Revista
Articulo
Indicadores
Citado por SciELO
Accesos
Links relacionados
Similares en SciELO
Compartir
Revista Técnica energía
versión On-line ISSN 2602-8492versión impresa ISSN 1390-5074
Resumen
CUESTA, A.; CONSTANTE, J. y JIJON, D.. Forecasting Models of Solar Radiation and Air Temperature through Recurrent Neural Network. Revista Técnica energía [online]. 2023, vol.19, n.2, pp.81-89. ISSN 2602-8492. https://doi.org/10.37116/revistaenergia.v19.n2.2023.552.
The aim of this study is to compare two architectures of recurrent neural networks of Elman and Jordan (RNRE and RNRJ), focus on the forecasting for two days of solar radiation and air temperature. The inputs of the forecasting model are meteorological variables as wind speed, atmospheric pressure, relative humidity and precipitation. The Research Institute for Geology and Energy of Ecuador provided the data of three meteorological stations situated in the provinces of Pichincha and Tungurahua for neural network training, validation and forecasting stages. Each network was trained with three different learning functions: backpropagation, backpropagation momentum and resilient propagation. The results shows the statistical parameters, Person correlation, mean square error and forecasting behavior on graphics for air temperature and solar radiation, according to RNRE and RNRJ model. This work shows correlation index greater than 0,9 in the validation stage. In the forecasting stage, the correlation index is higher than 0,8 and the mean square error shows values less than 0,02 kW for solar radiation and 2 ºC for air temperature.