多变量系统神经网络辨识的无模型自校正控制器研究  

Self-Tuning Controller with Model Free of Multivariable Neural Network System Identification NARMA Model

作　　者：侯小秋 Hou Xiaoqiu(School of Electronics and Controlling Engineering,Heilongjiang University of Science and Technology,Haerbin City,Heilongjiang Province 150022)

机构地区：[1]黑龙江科技大学电气与控制工程学院,黑龙江哈尔滨150022

出　　处：《黄河科技学院学报》2024年第5期1-5,共5页Journal of Huanghe S&T College

摘　　要：针对多变量NARMA模型,将其转化为具有耦合的子系统,采用具有辅助变量的多变量紧格式动态线性化方法逼近多变量NARMA模型,利用BP神经网络辨识其参数,基于多变量广义目标函数提出多变量NARMA模型的神经网络辨识的无模型自校正控制器,算法为关于控制输入的非线性方程组,通过非线性数值分析的牛顿法对其进行求解,根据非线性递推最小二乘法对BP神经网络的连接权重值进行在线学习。仿真研究表明系统的响应具有优良的性能。Regarding the multivariate NARMA model,it is transformed into subsystems with coupling.A multivariate tight-format dynamic linearization method with auxiliary variables is adopted to approximate the multivariate NARMA model,and BP neural network is utilized to identify its parameters.A self-tuning controller with model free of neural network identification of multivariable NARMA model is developed by basing on generalized multivariable object function,and the self-tuning controller is nonlinear equations with control input.The nonlinear equations are solved by Newton method of nonlinear numerical analysis.Online learning of connection weight value of BP neural network is conducted by nonlinearity recursive least squares method.Simulation results indicate that the system response has excellent performance.

关 键 词：神经网络控制 无模型自适应控制 自校正控制器 多变量非线性系统 多变量广义目标函数 牛顿法 非线性递推最小二乘法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]