A Learning Rate For MIMO Nonlinear System Emulation

This paper presents an emulation scheme based on a novel method to adjust the learning rate for multivariable nonlinear dynamical systems. The aim of this paper is to adapt the learning rate of the Neural Emulator (NE) in order to accelerate the convergence speed and to improve the precision degree. To ensure fast convergence and good estimation, an online adaptation is developed using a criterion generated by the error of emulation. The obtained results prouve the efficiency of the designed NE compared to those obtained with an existing one using a fuzzy supervision.