Proposição de modelo de predição de desempenho de estações convencionais de tratamento de água utilizando redes neurais artificiais

Abstract

Water treatment plants (WTP) have a fundamental and strategic role in the control of waterborne diseases through the potabilization of water, in order to meet the needs of the population that is supplied by it. In this context, evaluating the performance of these stations is paramount, particularly for the entities responsible for the water quality control stage, since WTP must present and operate with minimum conditions necessary to achieve its objective. As a way of anticipating adverse situations, as well as in the search to help decision makers and managers responsible for such operations, several models of WTPs performance prediction were proposed, making use of a series of tools, notably computational modeling, an important area of Artificial Intelligence (AI). As a contribution to this, this PhD thesis aimed to propose performance prediction models of conventional water treatment stations using Artificial Neural Network (ANN), with turbidity and apparent color of the treated water as control parameters. For its development (five ANN models were developed, two for small WTPs, two for medium-sized WTPs and one for large WTPs), data on raw and treated water quality were used. and hydraulic parameters of 12 water treatment plants, four with flows of up to 50 L.s-1, three with flows between 50 L.s-1 - 500 L.s-1 and five with flows above 500 L.s-1. The models were developed using RStudio®, from the artificial neural network multilayer backpropagation, with logistic activation function. The models presented determination coefficients of 0.66 and 0.67 for the turbidity and the apparent color of the treated water of small stations, respectively. The R² values were 0.21, 0.33 and 0.60, respectively, for the medium (Models 3 and 4) and large (Models 5). In the comparison of the results, models were also developed in multiple linear regression, being found that the models in ANN presented better performance. The results corroborate the application of AI techniques, particularly ANN, in WTPs, in order to optimize the process, reduce consumption of chemicals and generate waste, as well as to make the operation more efficient and safer by offering a product more reliable to the population supplied.