Nonlinear system identification using neural networks trained with natural gradient descent (Q1886938)

An unknown nonlinear system is assumed to be a blackbox and the learning process is performed using the input-output signal only. Natural gradient (NG) learning neural networks (NNs) are used for modeling nonlinear systems with memory. The NN model is composed of a linear adaptive filter followed by a two-layer memoryless nonlinear NN. A Kalman-filter-based technique and a search-and-converge method are employed for the NG algorithm. Computer simulations and illustrations are given.