Neural networks for long term prediction of fouling and backwash efficiency in ultrafiltration for drinking water production

The aim of this study was to develop a neural network model to predict the productivity of an ultrafiltration pilot plant, treating surface water to produce drinking water and operated with sequential backwashes. The model had to predict long-term performances of the pilot plant, it means to consider both reversible and irreversible fouling. The model had also to take into account a minimum number of parameters. On site experiments were performed to constitute the learning and validation databases. The developed model consists in two interconnected recurrent neural networks. It allows predicting satisfactorily the filtration performances of the experimental pilot plant for different resource water quality and changing operating conditions.