Identification of Systems Having Unstable Dynamics and Time Delays Using Delayed Recurrent Neural Networks.

This article addresses the challenging problem of identifying unstable system dynamics with time delay. In the proposed novel scheme, a recurrent neural network (RNN) with parallel delayed architecture in closed-loop has been employed, which is referred to as closed-loop-delayed-RNN (CLDRNN). The systematic mathematical formulation is done in easy to follow steps to calculate all the parameters of unstable delayed process models. Interestingly, in proposed algorithm, all model parameters are directly estimated in terms of the optimized CLDRNN weights only, without using any prior knowledge about the unknown process dynamics. The Lyapunov theory is incorporated to get efficient learning, and an accurate condition is derived to achieve guaranteed global convergence of the proposed algorithm. Various identification experiments are conducted on benchmark unstable process examples to show the efficacy of the proposed approach. [ABSTRACT FROM AUTHOR]