PRONÓSTICO DE LA MIGRACIÓN DE CONTAMINANTES EN AGUAS SUBTERRÁNEAS MEDIANTE REDES NEURONALES ARTIFICIALES

Abstract

Currently, it is reported the application of artificial neural networks (ANN) in areas of environment, water, medicine, among others. It was considered feasible to conduct an RNA model to predict the contaminant transport in saturated media, homogeneous and isotropic for the textural classes proposed by the Department of Agriculture of the United States. The models were trained and evaluated from the equation proposed by Ogata and Banks (1961), which considers the diffusive terms and advectivos. Required parameters for the models are: hydraulic conductivity (K), effective porosity (θe), diffusion coefficient (Dd), distance (x) and time (t). Developed structures of ANN backpropagation in multilayers. Neurons in the hidden layer were varied between 4 and 10, according to the rules proposed by Goethals et al., (2007). To train the ANN used the Levenberg-Marquardt algorithm, was applied the transfer function in the log-sigmoidal hidden layer and linear output layer. The choice of the model ANN was based on the generalization error was finally performed a statistical analysis using the correlation coefficient (R), the χ2 test (ji square) and Student t for the training phase and test, to ensure the 99% confidence level proper training and performance. This paper is divided into 5 chapters, the first entitled Environmental Issues Associated with the conservative Contaminants addresses the major problems of conservative pollutants such as POPs, heavy metals, pesticides, etc., And presents the same problems exist at different sites globally and nationally, in Chapter 2 Theoretical Framework addresses the phenomena that governs the movement of contaminants in groundwater, as well as the basics of artificial neural networks (ANN) in Chapter 3 Development model, where the modeling equation with the advection - dispersion, also, is the training and test models for the ANN textural classes proposed in Chapter 4, Evaluation of Results, we analyze the results obtained, and a statistical evaluation of models for ANN and finally in Chapter 5, Conclusions and Recommendations, presents the conclusions and recommendations reached in this thesis work.