Hybrid Model of Cooling Tower Based on First Principles and Neural Networks

Abstract

Mathematical modelling in paper industry is to a large extent based on physical models that usually contain empirical correlations and assumptions. Milosavljevic and Heikkilå [1] derived a mathematical model for a counter-flow wet cooling tower based on one-dimensional heat and mass balance equations using an expression for the volumetriv heat transfer coefficient. In the present study, volumetric heat transfer coefficient values determined by the above-mentioned model on the basis of measurements from a pilot cooling tower were approximated with feed-forward neural networks. It was found that a considerably more accurate approximation of the coefficient was obtained with the neural network than by conventional non-linear regression. The implementation of the expression in the first-principles model can lead to more accurate dimensioning of cooling towers, which results in lower equipment costs and better energy utilization.