Pronóstico de la concentración de ozono en Guadalajara-México usando redes neuronales artificiales

Abstract

A forecasting model to predict the maximum ozone concentration in a specific day in the Metropolitan Area of Guadalajara-Mexico was developed. An Artificial Neuronal Network fed with six meteorological variables and three chemicals was used. Nodes in the hidden layer were varying in a number among 12 and 15. The transfer functions were log-sigmoid for the hidden layer and linear for the output layer. For the network training the Levenberg-Marquardt algorithm with historical dates from 1999 to 2004. Data for 2005 were used to evaluate the predictive capabilities of the trained network, evaluating the quality of the air at three levels: good, moderate, and unhealthy. The model presented global efficiencies of around 50%, reaching and 65% for high ozone concentrations.