A memory ann computing structure for nonlinear systems emulation identification

Abstract

Currently, almost all efforts for using artificial neural networks for control oriented process identification are based on feed-forward networks. Provided the system order or the upper limit of the order is known, a neural network design is feasible for which all the collection of previous values of the inputs and outputs of the system to be identified can be used as input data to train in the network computing structures to learn the input-output map. This work reports on a novel technique that makes use of memory artificial neural network architecture that can learn and transform so as to emulate any non-linear input-output map for multi-input-multi-output systems when no prior knowledge on specific system features exists.