Automated neural network model selection algorithm for feedback linearization based control

Abstract

For the best model identification a set of neural networks (NNs) must be trained. First of all it is necessary to obtain the optimal structure of the NN. In addition a good choice of the initial values of the NN parameters can be of tremendous help in a successful control application. Further fit of the model is evaluated using several control criteria, and the optimal among them is selected. This article presents an automated NN model selection method for control based on feedback linearization.