Prediction of dynamic state and control for specification change by neural network for continuous bulk polymerization of polystyrene

Two simulators for bulk polymerization of polystyrene were made. One was used to represent the real reactor and the other was simpler and used as process model. Through them, neural networks were constructed for both dynamic state prediction and process control. Dynamic state prediction to predict the change of reaction temperature was achieved successfully by applying local and global learning simultaneously. With the former learning the actual experimental data were taught, while in the latter learning, global features over a wide range of operating conditions, that were given by the process model, were taught. Control was examined by two kinds of network. Both local and global learning were applid by a neural network, and the reaction temperature was controlled by manipulating either the coolant temperature or the feed rate. Although its calculation is simple and fast, satisfactory control was achieved. By using a forword network, the reaction temperature was manipulated by both the coolant temperature and the feed rate. By use of two operating variables, stabler central was obtained for wider target changes.