Coagulation is a stage in water treatment and, for this, jar tests are performed, which allows determining the optimal coagulant and alkalizer doses in coagulation process. However, these tests are time-consuming and do not enable real-time responses to changes in raw water quality. To overcome these limitations, artificial multilayer perceptron neural networks were built, trained, validated and tested to predict the aluminum and sodium hydroxide doses - used as coagulant and alkalizer, respectively. The results of these models are encouraging to consider that the estimated uncertainties have the same order of the variation limits magnitude indicated by the jar tests for almost a six-year period.
artificial neural networks; coagulation process; jar test; alum; sodium hydroxide
