Neural network - based estimation of reaction rates with partly unknown states and completely known kinetics coefficients

This work is focused on developing a more efficient computational scheme for estimation of process reaction rates based on NN models. In contrast to the traditional way of process reaction rates estimation by exhaustive and expensive search for the most appropriate parameterized structure, a neural network (NN) based procedure is proposed here to identify the reaction rates in the framework of an analytical process model. The reaction rates are not measured, therefore a special hybrid NN training structure and adaptation algorithm are proposed to make possible the supervised NN learning. The present contribution is focused on the general modelling of a class of nonlinear systems representing several industrial processes including crystallization and precipitation, polymerization reactors, distillation columns, biochemical fermentation and biological systems. The proposed algorithm is further applied for estimation of the sugar crystallization growth rate and compared with alternative solution.