Robust training algorithm of multilayered neural networks for identification of nonlinear dynamic systems

Motivated by adaptive control systems, a dead zone technique is used for the nonlinear gradient descent algorithm to train a multilayered feed-forward neural network to identify nonlinear dynamic systems. The dead zone scheme guarantees convergence of the neural network in the presence of noise. Simulation results are presented to demonstrate the robustness of the algorithm. A local convergence proof of the robust training algorithm is also provided.