基于神经网络的非线性系统的观测器设计

Neural-network-based observers for nonlinear systems

非线性系统的观测器设计无论在理论上还是在实际应用中 ,一直都是控制界研究的重要课题之一。针对一类非线性系统 ,论文用前馈神经网络的函数逼近能力 ,提出了基于神经网络的观测器设计方法 ,建立了 L yapunov函数 ,并给出了网络权系数矩阵的在线学习规则为δ-修正。证明了网络权矩阵是最终一致有界的。最后针对单臂机器手实例 ,给出了仿真实验。

Design of observers in nonlinear systems is an important past of control research. The functional approach of forward feed back neural networks provides a theoretical and practical design method for neural networks based observers for nonlinear systems and for the creation of a Lyapunov function. This method which handles multi variable dynamic systems with unknown nonlinearities to shown theoretically to be the correct and effective. δ modification of on line learning rules were given for the network weight matrix which is proven to be ultimate unanimousl bounded. Results are presented for a one arm robot simulation.