Modelagem e implementação de um sistema inteligente de apoio operacional aplicado ao processo de tratamento de água

Abstract

Water is an essential resource for life to be maintained. However, to be appropriate to the consumption of humans, the water quality must meet several potability requirements set by legislation. It´s largely abstracted from surface water resources, such as dams and rivers, and transferred to the water treatment process. Among the various existing chemical processes in the treatment there is the coagulation, which has the purpose of destabilizing the electric charges of the undesirable substances causing the formation of flakes that will be retained in the treatment process. The coagulant dosage is defined generally using the methods of jar-tests and measurements of zeta potential, which are required at each change of quality of raw water (water in the spring characteristic). Jar-test tests require human resources and have the time of implementation and potential long analyzers zeta has high cost of acquisition. The results of both procedures are further improved during use in water treatment process, according to the experience of the technician treatment. The techniques of Computational Intelligence, such as Artificial Neural Networks had been used to provide optimized values of coagulant dosage on quality change in times of raw water and the turbidity of the filtered water, aimed at cutting costs in the process with chemicals and human intervention. The simulations were performed with historical data applied to the models studied with topologies 25, 40, 60 and 80 neurons in the intermediate layer. The topology with 25 neurons in the intermediate layer showed better results in the simulations and was implemented in real time. The results indicate reduction of inputs and encouraging the replacement of methods that require human resources, and high cost equipment.