Predicción de caudales máximos diarios del Río Magdalena a la altura del municipio de Calamar (Bolívar), usando el método de redes neuronales artificiales (Rna).

Abstract

In hydrology, it is usual to use traditional physical models to determine hydro-climatic factors; however, these have presented several limitations, among them practicality and precision due to the non-linear behavior of natural phenomena. In particular, the projection in time of hydro-meteorological variables such as flows and / or levels has become an imperative need for the optimal management of water resources; specifically regarding the generation of electrical energy; territorial ordering; design and operation of structures; supply and prevention of emergencies. The recent digital evolution and improvement of computer models, has generated an increase in the use of technological tools in various areas of science. Hydrosciences, for example, to solve physical problems and approximate unknown magnitudes, have required unconventional prediction methodologies based on artificial intelligence (AI), considering within this discipline artificial neural networks (ANN) as powerful in the capacity for learning and adaptation. This work exposes the implementation of five RNA models in Matlab®2020a, for the prediction of maximum daily flows [Q_MX_D5] of the Magdalena River, using as input attributes the day of the year (DOY, for its name in English), the pluviometric day [PTPM_CON6] and daily maximum level [NV_MX_D7] for different time series. The analysis of the results was carried out from the mean square error (MSE) and the correlation coefficient (R). The most optimal RNA model and closest to the real data was, RNA_4_1 with MSE of 3,882E-07 and R equal to 1. The study was carried out in a small section of the Magdalena River, between the SAN PEDRITO ALERTA and CALAMAR stations. The data was take from the information bank of the Institute of Hydrology, Meteorology and Environmental Studies (IDEAM) from January 2017 to December 2020.