基于SCFNN的网络控制系统远程控制器设计  

Remote Controller Design of Networked Control Systems Based on Self-Constructing Fuzzy Neural Network

作　　者：李毅[1] 彭勤科[1] 胡保生[1] 

机构地区：[1]西安交通大学系统工程研究所机械制造系统工程国家重点实验室,陕西西安710049

出　　处：《系统仿真学报》2005年第6期1307-1310,1314,共5页Journal of System Simulation

基　　金：国家自然科学基金(60175015和60373017)

摘　　要：提出将自组织模糊神经网络(SCFNN)应用于网络控制系统(NCS)中的远程控制器的设计。SCFNN的学习过程包括结构学习和参数学习两个阶段。结构学习的目的是对输入空间进行合理的模糊划分并动态地生成一组模糊逻辑控制规则,而参数学习是通过有监督梯度下降法来调整隶属度函数的参数以及模糊规则中结论部分的权值。最初的SCFNN只有输入节点和输出节点,而经过在线学习后逐步生成隶属度函数节点和规则节点。对基于Profibus-DP网络的网络控制系统进行测试,并与采用修正Ziegler-Nichols法设计的控制器的控制效果进行比较,结果表明基于SCFNN思想设计的远程控制器在网络控制系统中能够获得满意的控制效果。A self-constructing fuzzy neural network (SCFNN) is proposed to design remote controller in networked control systems (NCSs). The structure and parameter learning phases are preformed concurrently and online in the SCFNN. The structure learning is used to obtain a set of fuzzy logic control rules and a proper fuzzy partition of input space, while the parameter learning is used to adjust parameters of the membership function and weights of the consequent part of the fuzzy rules based on the supervised gradient descent method. The initial SCFNN consists of input and output nodes only. In the learning process the nodes of the middle layers, which correspond to the membership functions and the fuzzy rules, are created gradually, so a set of fuzzy rules is achieved dynamically. Numerical results on a test system using Profibus-DP network are presented and compared with those of the modified Ziegler-Nichols method. The results show the effectiveness of SCFNN in designing remote controller for NCSs.

关 键 词：自组织模糊神经网络(SCFNN) 网络控制系统(Ncs) 结构学习 参数学习 有监督梯度下降法 

分 类 号：TP273.5[自动化与计算机技术—检测技术与自动化装置]