Coagulant dosage prediction in the water treatment process

Abstract

Coagulation is an important water treatment step in a water treatment plant (WTP). Jar tests are performed to determine the required dose of coagulant; however, these tests are slow to be performed and do not give a response in real-time to changes in raw water quality that changes abruptly during the day. To overcome this limitation, this research developed artificial neural network (ANN) models, using full-scale WTP data that served to calibrate the model and then predict the coagulant dosage, considering raw water as data input, in compliance with the treated water quality parameters. The best model was able to predict the coagulant dosage with a mean squared error of 0.016 and a correlation coefficient equal to 0.872. These results corroborate to promote coagulant dosage automation in WTPs, making it clear that ANN models allow a faster response in dosage definition and reduce the need for human interaction in the process.