一种基于动态最近邻聚类算法RBF网络非线性系统复合控制器设计

The design of a multiplexed controller for use with a nonlinear system based on dynamic nearest neighbor-clustering algorithm for RBF neural networks

针对RBF网络的设计难点,提出一种动态确定隐层节点数和聚类中心的新方法。并基于逆动力学的思想,提出一种RBF网络逆控制与PID控制相结合的在线自学习控制方案。辨识器采用RBF网络结构和动态最近邻聚类算法,实现了对系统逆动力学的动态辨识。并将辨识模型作为控制器模型,与被控对象串联,构成一个动态伪线性系统,从而使非线性对象的控制问题简化成线性对象的问题。仿真结果证明了该控制策略具有良好的动态跟踪性能和抗干扰能力,具有较强的鲁棒性。

This paper presents a method of controlling the RBF neural network data center of a hidden layer based on the design feature of the RBFNN, and makes a proposal for an on - line self - learning control strategy in the light of the thought of inverse system control, which combines the RBFNN - based inverse control with PID control. The system identifier makes use of the RBFNN structure and the dynamic nearest neighbor - clustering algorithm to implement the identification of an inverse dynamic system model. The controller model is connected in series with the plant, thus forming a dynamic pseudo linear system. Consequently, the control problem of a nonlinear plant is converted to a linear system. The result from simulation shows that the control strategy can provide formance and resistance to disturbance, but also with the system not only with the nice dynamic track pergreat robustness.