IMPROVED OPERATION OF A BATCH POLYMERISATION REACTOR THROUGH BATCH-TO-BATCH ITERATIVE OPTIMISATION

摘要

A batch-to-batch iterative product quality optimisation control strategy for a batch polymerisation reactor is proposed. Recurrent neural networks are used to model the dynamic behaviour of product quality variables. Model-plant mismatches and unknown disturbances are reflected in the model prediction errors. The repetitive nature of batch processes enables this information being discovered from previous batches and used to improve the current batch operation. Recurrent neural network predictions for the current batch are modified using prediction errors in previous batches. Because modified model errors are gradually reduced from batch to batch, the control trajectory gradually approaches to the optimal control policy and tracking errors also converge. The proposed scheme is illustrated on a simulated batch polymerisation reactor.