MODELING OF COAGULANT DOSAGE IN A WATER-TREATMENT PLANT

The coagulation-flocculation is a major step in the drinkable water tr eatment process allowing the removal of colloidal particles. The water treatment facilities of the City of Sainte-Foy have been well instrum ented and process variables such as temperature, pH, turbidity, conduc tivity of raw and treated water along with actual coagulant dosage ava ilable have been measured and stored each 5 min for several years. Usi ng such a data bank, the objective of this paper is to report on the d evelopment of a neural network predictor of coagulant dosage in order to facilitate process operation. Feedforward neural models have been b uilt using a quasi-Newton method along with the early stopping approac h to avoid overfitting. Annual and seasonal models have been built and their performances are discussed. (C) 1997 Elsevier Science Limited.