Fernanda Cristina Arantes,Alex Almeida Magalhães,Alexandre Branquinho Rocha,Aladir Horácio dos Santos,Christiane Pereira Rocha, Unifor-MG, Engenharia Ambiental, rochachristiane@uol.com.br
ABSTRACT
The objective this study was to develop a model that provides a reliable prediction of BOD input and output of artesian wells at Formiga-MG. Stationary and dynamic models were developed using the technique of artificial neural networks and measures of water quality such as BOD, COD and DO. The obtained results are in agreement with standard reference ranges except for COD, probably because there is a stream near the artesian well. The neural network showed a strong correlation (R = 1) when compared to other methods, however the learning curve showed no uniformity in the results, generating errors when it tried to predict new results, a fact that probably occurred because the network has been trained only with data of COD and BOD. One of the conclusions obtained is that besides the COD other factors can affect the response of the BOD, such as pH, DO and turbidity.
Palavras-chaves: Groundwater, BOD, Artificial Neural Networks