AWWA MTC53944

The aim of this work was to develop a model to predict long-term performances ofan ultrafiltration pilot plant producing drinking water from surface raw water.On-site experiments were performed to form learning and validation databases. Themodel consists of two interconnected recurrent neural networks coupled by Darcy'slaw. The parameters taken into account are water quality parameters necessary torepresent fouling, operating conditions during filtration time and during thebackwash procedure to predict its efficiency. The model satisfactorily predictsthe filtration performances of the experimental pilot plant for different waterquality and changing operating conditions, during several hundred filtrationcycles. The model was then used to predict the influence of some operatingconditions on irreversible fouling for a given surface water. New criticaloperating parameters were then introduced which are adapted to a filtrationprocess operated with sequential backwashes and which can provide operatingguidelines to prevent irreversible fouling and to limit the use of chemicalcleanings. Includes 10 references, tables, figures.