Process modeling and control using a single neural network

Artificial neural networks provide a useful tool for analyzing many industrial formulation, pattern recognition, and process control problems. The back propagation network [BPN] with a single hidden layer can map inputs to outputs for arbitrarily complex nonlinear systems, static or dynamic. The same BPN model can be pseudo-inverted, as described here, to determine the input variable set-points required to produce a specified pattern of output responses. Thus a single neural network model can be used for closed-loop supervisory process control. In a textbook example, a simple heat exchanger, this neural network methodology generates an optimal control strategy