A novel online black-box identification method for unknown nonlinear system via differential neural network

作者：Fu Zhijun^*; Ning Xiaobin; Jiang Qingjian

来源：International Journal of Innovative Computing Information and Control, 2017, 13(2): 471-484.

• Black boxes
• Differential neural network
• Nonlinear identifier
• Robust property
• Unknown nonlinear systems

In this paper, two differential neural networks (DNN)-based adaptive identifiers for unknown nonlinear systems are proposed. The first proposed identifier is with the single layer DNN and the second one is with the multilayer DNN. Lyapunov approach is used to develop the online updating laws for the dynamic linear matrix arid the weights of the proposed two DNN identifiers. Moreover, robust proper ties of the proposed two DNN identifier's are proved by means of passivity approach, and the commonly used robust modification methods such as dead-zone, e-modification or σ-modification are not needed. Simulation results of an engine idling system demonstrate that the proposed identifier with the multilayer'DNN is more accurate than the proposed identifier with the single layer'DNN, and both of them illustrate improved performance compared to the conventional neural network-based identifier based on the assumption that the linear'matrix is known a priori.

• 出版日期2017
• 单位浙江大学