Robust identification for unknown nonlinear multivariable systems based on dynamic neural networks

Qionghai Dai, Tao Zhang, Yu-mei Zhang, Tianyou Chai, Li-hua Xia

Abstract: In this paper a learning and identification scheme for a class of unknown multivariable nonlinear system using dynamic neural networks (DNN) is presented. A DNN identifier is employed to perform "black box" identification. The identification scheme, based on DNN model, is then developed using Lyapunov synthesis approach with the projection modification method. The feature of this approach is that neither off-line learning phase nor all plant states for measurement are required. It is shown theoretically that the identified system is robust stable and the identified error is ensured in a stable region with respect to modeling errors and unmodeled dynamics. Simulation results with unknown nonlinearities are given to demonstrate the effectiveness of the proposed identification algorithm.