Identification of a class of hybrid dynamic systems with feed-forward neural networks: About the validity of the global model

Abstract

This paper addresses the problem of the identification of a class of Hybrid Dynamic System (HDS). The class herein considered is characterised by continuous inputs, continuous outputs and binary discrete inputs. The proposed approach focuses the attention on the identification of a global model that predicts the continuous outputs of the HDS. The accuracy of this global model is discussed and several simulation examples are investigated in order to study the validity of the obtained neural network global model. The originality of this approach consists in the identification of a HDS without needing to cluster the data or to know the current mode because it considers the identification of HDS in terms of the architectures and the learning algorithms developed for Feed-Forward neural networks.