Modelling of methanol synthesis in a network of forced unsteady-state ring reactors by artificial neural networks for control purposes

A numerical model based on artificial neural networks (ANN) was developed to simulate the dynamic behaviour of a three reactors network (or ring reactor), with periodic change of the feed position, when low-pressure methanol synthesis is carried out. A multilayer, feedforward, fully connected ANN was designed and the history stack adaptation algorithm was implemented and tested with quite good results both in terms of model identification and learning rates. The influence of the ANN parameters was addressed, leading to simple guidelines for the selection of their values. A detailed model was used to generate the patterns adopted for the learning and testing phases. The simplified model was finalised to develop a model predictive control scheme in order to maximise methanol yield and to fulfil process constraints.