Resumen

CERVANTES-OSORNIO, Rocio; ARTEAGA RAMIREZ, Ramón; VAZQUEZ PENA, Mario A.  y  OJEDA BUSTAMANTE, Waldo. Backpropagation artificial neural network versus empirical models for estimating daily global radiation in Sinaloa, Mexico. Rev. Mex. Cienc. Agríc [online]. 2016, vol.7, n.5, pp.1029-1042. ISSN 2007-0934.

The results were compared of average daily global radiation model estimated with artificial neural network (RNA) backpropagation against those obtained by empirical models Hargreaves, Angström-Prescott and these calibrated. A model of backpropagation artificial neural network was used with Levenberg Marquardt algorithm for forecasting average daily global radiation four stations located in the irrigation district 075 Valle del Fuerte, Los Mochis Sinaloa, Mexico. The database represents daily averages with 1 484 data vectors for training, validation and test and 229 for prognosis. Among the input variables provided by the irrigation district they were: minimum temperature and maximum temperature, others were calculated as actual duration of sunshine, photoperiod and extraterrestrial solar radiation. The scenarios with one, two and three hidden layers with different numbers of neurons in each hidden layer was obtained. The RNA e6{27} with entries minimum temperature, maximum, hours shine sun divided by photoperiod and extraterrestrial solar radiation, obtained the best fit with a RMSE of 1.6871 and R2 of 0.89 for 1 484 and for data for 229, the AngstromPrescott won the calibrated model with RMSE of 2.2812 and R2 of 0.89. For 1484 average data, the e6{27} scenario presents the best estimate of daily global radiation (Rs) and is better than the empirical models, however for 229 data the Angstrom-Prescott calibrated model provides an estimate of Rs better e6{27} of the RNA.

Palabras llave : Angström-Prescott; Hargreaves; averages; artificial neural network; solar radiation.