基于神经网络的非线性多模型自适应控制  

Nonlinear Multi-model Adaptive Control Based on Neural Networks

作　　者：姚健[1,2] 纪志成[1,2] 黄言平[3] 

机构地区：[1]轻工过程先进控制教育部重点实验室,江苏无锡214122 [2]江南大学物联网工程学院,江苏无锡214122 [3]北京华航无线电测量研究所,北京100013

出　　处：《控制工程》2014年第2期172-177,共6页Control Engineering of China

基　　金：国家自然科学基金项目(61203111)

摘　　要：针对一类非线性离散动态系统,设计了一个自适应控制方案。为了保证在任意时刻均能为被控的动态系统选择最好的控制器,方案基于输入输出数据为系统定义一个线性预测模型,并在此基础上设计能够保证闭环系统所有信号有界的线性鲁棒自适应控制器,同时定义一个非线性预测模型,再基于径向基神经网络设计一个旨在提高系统控制性能的非线性自适应控制器。通过比较2个控制器预测的系统输出性能,设计合理的开关切换规则。控制方案能将系统稳定性控制和性能优化的控制分离并单独实现,使得系统能在保证稳定性前提下,借助神经网络控制器良好的追踪能力有效提高自适应控制效果。最后通过仿真例子说明了系统稳定和提高输出追踪效果可以同时得到保证。For a class of discrete time dynamical systems, a new adaptive control scheme is proposed in this paper. In order to guaran- tee a best choice for the system is selected at anytime, a linear prediction model is defined based on input-output data and a linear ro- bust adaptive controller is designed accordingly to hold all the signals bounded; meanwhile, a nonlinear prediction model is defined for the designing of a radial basis function based neural network adaptive controller, the combining a linear robust adaptive controller and a neural network based nonlinear adaptive controller to improve the control performance. By comparing the predicted output performance of these two models, a switching law is well designed. The control of stability and performance improving can achieve respectively, which not only guarantees the stability, but also improves the adaptive control performance by using neural network controller. Finally, it is demonstrated that improved performance and stability can be simultaneously achieved by simulation examples.

关 键 词：非线性系统 自适应控制 神经网络 多模型 开关转换 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]