MODELO PRONÓSTICO PARA CONCENTRACIONES DE CLOROFILA, EMPLEANDO REDES NEURONALES ARTIFICIALES.

Abstract

Two Artificial Neural Networks (ANN) models were developed for chlorophyll concentrations prediction in a water body; both models were based on a Retropropagation ANN structure. The MATLAB™ program was used to develop the models. The forecast models were, for the Dam of Valle de Bravo, located in the State of Mexico, Mexico and the second for Yuriria Lagoon, located in the State of Guanajuato. Historical data were taken from parameter measured which determined water quality in the period of 1998-1999. Models were trained with eleven physicochemical variables from six sampling stations in the case of Valle de Bravo and nineteen variables collected at five sampling stations in the case of Yuriria, and tested with chlorophyll concentrations selected at random from the database that was used during the training. The Nodes that the models had in the hidden layer were varied between 10 and 30 for both cases. The hidden layer transfer functions were log-sigmoidal, and the output layer was linear. The Levenberg-Marquardt algorithm was used in the training phase. Lastly, a statistical analysis was made for the training and test phases, which allowed ensure 99 % level of confidence. Also the rate of eutrophication was calculated.