Modelling of a paper making process via genetic neural networks and first principle approaches

Presents a novel approach for the modelling of typical nonlinear systems at the wet end of the paper machines. Due to the complicated nature of the process, at first a pure multilayer perceptron (MLP) neural network is applied to establish the nonlinear model for the system. Instead of using standard backpropagation (BP) algorithms, genetic algorithm-based training is applied during the weight optimization phase. This is then followed by a logical combination of the so-formed neural network with a physical modelling exercise, leading to an improved semi-physical model which combines the advantages of physical and neural network modelling. The effectiveness of the proposed modelling techniques is illustrated by their applications in the establishment of two models, paper sizing and dry strength, for the wet-end chemical processes of the UMIST pilot paper machine.