A neural network strategy for batch process optimization

Abstract

The traditional way of operating a batch reactor is to use a pre-determined input trajectory. However, due to batch-to-batch variation in process conditions, this can lead to sub-optimal operation. In this paper, a novel neural network strategy is developed for calculating the optimal operating trajectory based on initial loading conditions and process parameters. The efficacy of this methodology is demonstrated via application to two industrially relevant batch polymerization processes — (1) batch styrene polymerization and (2) batch methyl methacrylate polymerization.