Model reduction for nonlinear DABNet models

We present an identification and reduction technique suitable for a particular nonlinear model structure. We use the DABNet structure, which is composed of a linear dynamic system followed by a nonlinear static map. The linear dynamic system is initially spanned by a set of discrete Laguerre systems, and then cascaded with a single hidden layer perceptron. A linear model reduction technique is performed on the hidden nodes of the neural network as part of the identification process. In that way, it is possible not only to identify the main time constants, but also to reduce the dimensionality of the perceptron input. Results concerning the application of the methodology to the approximation of a polymer process are presented