Desenvolvimento de modelos em redes neurais artificiais para a simulação da concentração do oxigênio dissolvido e da autodepuração do Rio Alegria - PR.

Abstract

The present study was to develop two models in Artificial Neural Networks (ANN) in order to estimate the dissolved oxygen and model the depuration of the Alegria River , located in the municipality of Medianeira in the State of Paraná. Both models were developed based on data of water quality of the river and model responsible for the prediction of self-purification of the river also used data from the effluent that is built into the river over the range studied With 22 samples in 6 seasons: the training and validation of models 132 sets of data were generated . To simulate the concentration of dissolved oxygen in the river water five different artificial neural networks have been developed. The input variables in these networks were the water quality parameters except the Dissolved Oxygen (DO), who set up in all the networks as output. To predict the self-purification of the river was a model developed in artificial neural networks based on water quality data at a point upstream of the release of effluent , this effluent and data , from these data, the model provides values of dissolved oxygen (DO ) and biochemical oxygen demand ( BOD ) at points downstream , ie, predicts the potential self-purification river . Among the models developed to simulate the dissolved oxygen concentration was the best result achieved an average error between the OD values estimated by the artificial neural network 11.4%. In the model tested to predict the selfpurification of the river, several architectures were tested and showed the best result was setup with a hidden layer of 14 neurons 14:14:1 and 2000 seasons. The best result of average error between the known values of BOD and the adjusted valueswas 2.17 % and the BOD generated an average error between the known values and the adjusted values of 11.26%.