A feed-forward artificial neural network as a simulator for a chemical process

Artificial neural networks (ANNs) can serve as simulators trained from observed gross behaviour of a process especially in cases where its mathematical modelling is qualitatively inadequate or not feasible. The inputs to the ANN may be various operating parameters, physical and chemical properties, and process conditions, while the outputs would be the final product variables characterising the output of the chemical process. The paper illustrates the feasibility of using a feed-forward neural network as a steady state simulator for chemical processes exemplified by a continuous stirred tank reactor with two first-order reactions in series, for which a simple mathematical model was used to generate a set of training examples. The network was trained by error square sum minimisation by the Levenberg-Marquardt method, and the influence of the number of hidden layers and the number of nodes in the hidden layers was studied