PROCESS IDENTIFICATION USING NEURAL NETWORKS

This report describes a series of tests in which a backpropagation neural network was used for process identification. The report demonstrates that improvements in neural network identification are possible by building upon established identification techniques. Modification of the network training set objective function and the use of cross validation are demonstrated in the report using both simulated and real process data. The report shows that the backpropagation neural network can learn a functional mapping between input and output based on a set of training examples. This capability is demonstrated for both static single-variable and multivariable systems and for linear and nonlinear dynamic systems. The report describes the tests that were performed and our conclusions. It further discusses the advantages and disadvantages of this technique in comparison with classical process identification techniques.