Partially unknown nonlinear systems identification based on two-neural-network

If the nonlinear system with unknown parameters is not in a controllable canonical form, then the derivative of the tracking error is unknown. The controller design for the system will be complex. In this paper, we propose two-neural-network controller (TNNC) to learn the unknown nonlinear coefficient of control systems. For the convenience of control, the structure of the two-neural-network is divided into two parts. Each part is a neural-network. Beginning, two neural-networks have the same structure. In the actual control process, one neural-networkpsilas output is used to approximate the output of controller, and the other neural-network is learning, the learning is the main on-line tracking learning. In control processing, the actions of two neural-networks can be exchanged by using the switching line, which can be gotten from Lyapunov energy function of the nonlinear system. Different with other neural network algorithms, here, the process of implement and learning of TNNC is divided, i.e., the learning of TNNC is fulfilled by one neural network, and the implement of TNNC is accomplished by another neural network. This does not make the learning to affect the real control. Stability analysis of this control law is also given in the paper.