Model identification of nonlinear time variant processes via artificial neural network

This paper demonstrates that neural networks in conjunction with recursive least squares can be used effectively for model identification of nonlinear time variant processes. The developed approach updates the process model partially at any given sampling time. By updating only a subset of parameters at a given time sample, rather than all network parameters, convergence time is significantly reduced. In addition, meeting the convergence criteria and over-parametrization are less of a problem. The updating approach is applied to a nonisothermal CSTR with time varying parameters and its performance is demonstrated. The resulting approach predicts the process output extremely well and has the ability to learn on-line. Copyright (C) 1996 Elsevier Science Ltd