Approximation Abilities of Locally Recurrent Networks

Contents

In the last decade, a growing interest in locally recurrent networks has been observed. This class of neural networks, due to their interesting properties, has been successfully applied to solve problems from different scientific and engineering areas. Cannas and co-workers [154] applied a locally recurrent network to train the attractors of Chua’s circuit, as a paradigm for studying chaos. The modelling of continuous polymerisation and neutralisation processes is reported in [155]. In turn, a three-layer locally recurrent neural network was succesfully applied to the control of non-linear systems in [132]. In the framework of fault diagnosis, the literature reports many applications, e.g. a fault diagnosis scheme to detect and diagnose a transient fault in a turbine waste gate of a diesel engine [109], an observer based fault detection and isolation system of a three-tank laboratory system [39], or model based fault diagnosis of sensor and actuator faults in a sugar evaporator [26]. Tsoi and Back [38] compared and applied different architectures of locally recurrent networks to the prediction of speech utterance. Finally, Campolucci and Piazza [156] elaborated an intristic stability control method for a locally recurrent network designed for signal processing.