A New Version of Elman Neural Networks for Dynamic Systems Modeling

Elman network is a class of recurrent neural networks used for function approximation. It has a set of global sigmoid functions at its hidden units. That means that if the operating conditions of a process be identified, are changed the function approximation property of the network is degraded. This is due to the fact that the universes of discourse of the network is covered by global sigmoid functions. This paper introduces a new version of the Elman network named Elman Recurrent Wavelet Neural Network (ERWNN). It merges the multi-resolution property of the wavelets and the learning capabilities of the Elman neural network to inherit the advantages of the two paradigms and to avoid their drawbacks. Structurally, the number of hidden/context units of an ERWNN should at least be equal to the order of the system be modeled. Stability and convergence property is proven for the proposed network. The dynamic back propagation (DBP) algorithm is employed to train the proposed network that can be used not only for modeling but also for control purposes. The advantages of this new version of ERWNN in modeling dynamic processes, can be reflected in our simulation results. Copyright © 2005 IFAC.