TIME SERIES MODELING OF ENVIRONMENTAL TEMPERATURE WITH ARTIFICIAL NEURAL NETWORKS, TINGO MARIA – HUANUCO MODELAMIENTO DE SERIES TEMPORALES DE LA TEMPERATURA AMBIENTAL CON REDES NEU

In this work, we model a time series of the average daily temperature for the city of Tingo Maria - Huanuco, using artificial neural networks. For the modeling of  daily temperature “output” was used as “inputs” network to the minimum and  maximum daily temperatures, hours of sun daily, extraterrestrial radiation, monthly climate index classification and grouping factor; was chosen to multilayer type network training algorithm "backpropagation" and a single hidden layer where thenumber of neurons from 1 to 12. varied Likewise periods of 30 and 74 years prior to knowledge extraction was used, with and without inclusion of periods where events occurred El Niño and La Niña, so it was expected that the models capture the nonlinear relationship between inputs and output. The results showed overall good performance of the models where a longer period for extracting knowledge (patterns 3 and 4 - 74 years previously) was held, resulting in a marked improvement in models where periods were not considered The boy and girl (pattern 4); the best model was made with the pattern 4, which obtained a Pearson correlation coefficient (r) of 0.9, a mean error (ME) 0; root mean squareerror (ECM) of 0.2 and a root mean square error (RMSE) of 0.5, showing that the simulated values to the target values (real temperature) correspond very well.