Red Neuronal Artificial y series de Fourier para pronóstico de temperaturas en el Distrito de Riego 075 Sinaloa México - Artificial Neural Network and Fourier series to forecasting temperatures of Irrigation District 075 Sinaloa Mexico

Abstract

Temperature is a transcendental variable in aspects such as evapotranspiration calculation, growth, development and yield of plants, in the study of the transmission pests and diseases, in the weather forecast, in determination of heat fluxes, in the calculation of the real vapor pressure, all these processes affected by global warming. The objective of this work was to compare the best results of two models: one of artificial neural network (RNA) backpropagation, and another of Fourier series. Daily data of maximum temperatures (Tmax) and minimum (Tmin) of the Santa Rosa 1, Ruíz Cortínez, Batequis and Santa Rosa 2 stations, of the Irrigation District 075 Valle del Fuerte, Los Mochis, Sinaloa, Mexico were used. In RNA, 1484 data vectors were used for training, validation and testing and 229 to forecasting. For training, the input variables of the RNA were: Julian day, longitude, latitude and altitude. Were obtained 96 scenarios with one, two and three hidden layers, with different numbers of neurons in each hidden layer. With the 1484 data, the best adjustments were obtained for the Fourier series models for maximum and minimum temperatures, and 229 data were predicted for the four stations. The best RNA backpropagation models for the prediction of maximum and minimum daily temperatures obtained similar performances in comparison with those made by the best models of Fourier series, for the study stations.

Keywords

• forecasting, artificial neural network, maximum temperature, minimum temperature