Modeling a scrubber using feed-forward neural networks

Mathematical modeling in the pulp and paper industry is largely based on ph ysical models that contain many empirical correlations and assumptions. A t echnique that can complement these physical models, and that does not requi re simplifying assumptions, is a neural network. Neural networks can be use d to model systems where physical models are not available or to improve ex isting physical models with various techniques of hybrid modeling. Such mod els have an advantage of the robustness of physical models combined with th e higher accuracies offered by empirical techniques such as neural networks . Most systems in paper drying are quite nonlinear and are often too complica ted to be accurately described with physical models. Neural networks are po werful tools that can solve a variety of nonlinear modeling problems. Using an example of a scrubber, this paper illustrates the versatility of this t echnique. Neural networks were used to predict the optimum operational cond itions in designing a scrubber for heat recovery from paper machines. In th is study, feed-forward neural networks are applied to estimate the outlet w ater temperature of the scrubber process. The networks are trained with dat a obtained from experiments carried out on two pilot scrubbers. The accurac y of the neural network model is significantly higher than that of the phys ical model.