I. INTRODUCTION
Adaptive control scheme based on recurrent neural networks seems a promising solution for uncertain or unknown nonlinear and multi-variable dynamical systems [1], [2], [3]. Thanks to their approximation capabilities, neural networks have been successfully applied to nonlinear plants control [4], [6], [5]. Indeed, the development of neural controllers requires efficient adaptation algorithms. The most popular algorithms used for parameters adaptation are the gradient back propagation learning algorithm and numerous variants [7], [8]. However, the gradient descent method is known for its slowness and its frequent confinement to local minima. So, it is very difficult to achieve stable and online controls in presence of unknown external disturbances and parameters perturbations.