It is common everywhere in the world, the necessity to know with enough time of advance the future flows in rivers where big cites and industries are settled. There are many methodologies that allow solving this problem, each one with their advantages and disadvantages. The ensemble and the comparison among diferent models of prediction is fundamental when analyzing the future situation in case of alert where it is necessary to make important decisions. In this thesis it has been carried out an intense bibliographical revision on the use of Artificial Neural Networks (RNA) to make flows predictions. This is made to know the state of the art of this methodology and starting from that point, propose and study improvements that can contribute to the advance of this art.
With the intention of giving physical meaning to this type of models, it has been proposed a hybrid model's methodology that allows identifying, automatically, the hydrological state of a certain basin, allowing to model separately each state using simple RNA. It has also been incorporated the physical concept in the election of the input variables, proposing geomorphologic and time responses of the basin that help to identify the most influential variables.
On the other hand, given the necessity to know the distribution’s function of the predictions in real cases, where it is necessary to make decisions from these results, it has been proposed a methodology to calculate the uncertainty of the predictions of any type of model without caring about their complexity.
To confer a practical use to these ideas, a computer application has been developed (ANN) which is able to carry out the necessary calculations for the construction of a forecasting RNA model. It incorporates modules for the analysis of the data, of the basins and river’s respond times, modules of parameter’s calculations, a module for the calculation of the uncertainty and others for the analysis of the results. This software has been created focused on the use in hydrology, since at the moment there are not programs that contain all the necessary elements and the flexibility oriented to the flows forecast. It has been also developed a guide of steps to follow for the obtaining of the optimal model, always keeping in mind the incorporation of the physical meaning of the process to model.