Fast and robust on-line system identification based on multi-layer recurrent neural networks

Abstract

Online system identification plays a crucial role in the adaptive scheme with unknown process dynamics. However, most systems have nonlinear, coupled and time-varying dynamics with uncertainties in practice. Neural networks provide such adaptive nonlinear models and can be well extended to system identification or prediction problems. In this research, the fast and robust online identification using multilayer recurrent neural network is tackled by two steps: (a) robust training method through modified recursive least-squares algorithm having dynamic forgetting factor. (b) multilayer network architecture through output and error recurrent neural network (OERNN).