Parametric identification using multilayer perceptron

In this work, we present an algorithm to extract a parametric model from a multi layer perceptron (MLP). The MLP is trained with input-output data collected from a process. This algorithm takes and processes the weights of the MLP to obtain a parametric model in difference equation form. As a first approach we use the MLP as a linear regresser. Next, we propose a scheme to extract the parameters from a neural network with non-linear activation functions applying Bayesian training. Comparisons between our parametric models and those using the identification system method show that the obtained models are similar to the traditional ones