Artificial neural networks (ANNs) have been applied to an increasing number of real-world problems of considerable complexity. Considered good pattern recognition engines, they offer ideal solutions to a variety of problems such as prediction and modelling where the industrial processes are highly complex. The present paper reports on the elaboration and the validation of a 'software sensor' using ANNs for online prediction of optimal coagulant dosage from raw water quality measurements, in a drinking water treatment plant. In the first part, the main parameters affecting the coagulant dosage are determined using a Principal Component Analysis. A brief description of this statistical study is given and experimental results are included. The second part of this work is dedicated to the development of a neural software sensor and the generation of an uncertainty indicator attached to the prediction. Bootstrap sampling has been used to generate a confidence interval for the model outputs. The ANN model was developed using the Levenberg Á/Marquardt method in combination with 'weight decay' regularization to avoid over-fitting. A linear regression model has also been developed for comparison with the ANN model. Experimental and performance results obtained from real data are presented and discussed.

A hybrid model for an anaerobic digestion process is proposed. The fermentation is assumed to be performed in two steps, acidogenesis and methanogenesis, by two bacterial populations. The model is based on mass balance equations, and the bacterial growth rates are represented by neural networks. In order to guarantee the biological meaning of the hybrid model (positivity of the concentrations, boundedness, saturation or inhibition of the growth rates) outside the training data set, a method that imposes constraints in the neural network is proposed. The method is applied to experimental data from a fixed bed reactor.

International audienceWe propose a general methodology to develop a hybrid neural model for a wide range of biotechnological processes. The hybrid neural modelling approach combines the flexibility of a neural network representation of unknown process kinetics with a global mass-balance based process description. The hybrid model is built in such a way that its trajectories keep their physical and biological meaning (mass balance, positivity of the concentrations, boundness, saturation or inhibition of kinetics) even far from the identification data conditions. We examine the constraints (a priori knowledge) that must be satisfied by the model and that provide additional conditions to be imposed on the neural network. We illustrate our approach with various biotechnological processes showing how to select the appropriate neural network architecture. The method is detailed for modelling an anaerobic wastewater treatment bioreactor using experimental data

No abstract

No abstract