UTILISATION D’UN MODELE HYBRIDE BASE SUR LA RLMS ET LES RNA-PMC POUR LA PREDICTION DES PARAMETRES INDICATEURS DE LA QUALITE DES EAUX SOUTERRAINES CAS DE LA NAPPE DE SOUSS-MASSA- MAROC

Abstract

This work describes a new approach to the prediction of the parameters (microbiological, physical-chemical) groundwater quality indicators in the water table of Souss-Massa Morocco. The originality of this work lies in the application of a hybrid model based on the Stepwise Multiple Linear Regression and Neural Networks Multilayer Perceptron type. During the first stage, conventional statistical models namely the Stepwise Multiple Linear Regression was applied to a database that consists of eleven vectors as input vectors of the model and three vectors as the model output vectors in order to optimize the explanatory variables. In a second step, the optimized data base in the first step was used to construct a non recurring multi-layer network, the weights of the network connections are determined using the gradient back propagation algorithm. The data used as a database (learning, testing and validation) of the hybrid model are those relating to the analysis of 52 groundwater samples collected at several stations distributed in space and in time, of the groundwater Souss-Massa Morocco. The dependent variables (to explain or predict), which are three in number, are the Electrical Conductivity EC, the amount of Fecal Coliforms CF and Organic Matter MO.