通用神经网络非线性系统模型参考自适应控制

General nonlinear system MRAC based on neural network

针对任意复杂的具有最小相位,滞后环节和非最小相位特性的离散非线性系统,提出一种通用的直接神经网络模型参考自适应控制。采用具有在线学习功能的最近邻聚类算法训练RBF神经网络控制器,同时引入优化策略对聚类半径进行自动调整,并利用构造伪系统的方法构成一种对非最小相位同样有效的神经网络模型参考自适应控制器。仿真研究证明,该控制策略不仅能使多种非线性对象跟踪多种参考信号,而且抗干扰能力和鲁棒性也很好。

This paper proposed a novel scheme of neural network model reference control for arbitrary complex nonlinear system, which in other words was the discrete time of minimum phase or lag or non-minimum phase systems. The method used the nearest neighbors clustering algorithm to train the RBF neural network（RBFNN）. Introduced the optimize strategy which regulated the clustering radius automatically to guarantee the rationality of radius. Through constructing pseudo-plant, put forward a neural network model reference adaptive control which was still effective to the nonlinear non-minimum phase. With the help of simulation, the control strategy based on RBFNN model reference adaptive control can not only make manifold nonlinear objects track multi-reference signals, but also possess resistance to disturbance and excellent robustness.