PREDICTION OF POLYMER QUALITY IN BATCH POLYMERIZATION REACTORS USING ROBUST NEURAL NETWORKS

A technique for predicting polymer quality in batch polymerisation rea ctors using robust neural networks is proposed in this paper. Robust n eural networks are used to learn the relationship between batch recipe s and the trajectories of polymer quality variables in batch polymeris ation reactors. The robust neural networks are obtained by stacking mu ltiple nonperfect neural networks which are developed based on the boo tstrap re-samples of the original training data. Neural network genera lisation capability can be improved by combining several neural networ ks and neural network prediction confidence bounds can also be calcula ted based on the bootstrap technique, A main factor affecting predicti on accuracy is reactive impurities which commonly exist in industrial polymerisation reactors. The amount of reactive impurities is estimate d on-line during the initial stage of polymerisation using another neu ral network. From the estimated amount of reactive impurities, the eff ective batch initial condition can be worked out. Accurate predictions of polymer quality variables can then be obtained from the effective batch initial conditions. The technique can be used to design optimal batch recipes and to monitor polymerisation processes. The proposed te chniques are applied to the simulation studies of a batch methylmethac rylate polymerisation reactor. (C) 1998 Elsevier Science S.A. All righ ts reserved.