Development of a neural sensor for on-line prediction of coagulant dosage in a potable water treatment plant in the way of its diagnosis

Coagulation is one of the most important stages in surface water treatment, allowing for the removal of colloidal particles. Its control, in the majority of the plants remains still manual and requires long and expensive analyses of laboratory which provide only periodic information. The present work describes an innovative methodology integrating various techniques for the development of a diagnosis system of this plant. A first part, concerned by this paper, consisted in developing a software sensor based on artificial neural networks for predicting on-line the amount of optimal coagulant dosage from raw water characteristics such as turbidity, pH, temperature, etc. In a second part this information will be integrated like an input to a diagnosis system of the plant. The development of the neural sensor has been performed based in real data covering several years of operation and its performance have been compared toward those given by a multi-linear interpolation.