Modeling of MSMPR crystallizer dynamics – time series prediction by neural network

Abstract

Mass crystallization process usually produces difficult for modeling oscillations of process parameters of diversified amplitude and period. For the simulation of dynamic behavior of MSMPR crystallizer in various technological conditions an artificial neural network specialized in time series prediction was originally used. The Monte Carlo simulations provided numerical data matrixes corresponded to stable and unstable process behavior, which were directly used for the neural network training and testing. Artificial neural network structures designed for both cumulative and individual parameter predictions were tested and verified in respect of their prediction ability in one-step, medium-term and long-term prognosis of the mass crystallization process dynamics.