多变量无模型预测神经网络线性自抗扰控制

Multivariable Model-free Predictive Neural Network Active Disturbance Rejection Linear Control

下载PDF

导出

摘要对难以建模的多变量非线性系统的控制难题,提出改进的具有辅助向量的多变量全格式动态线性化方法,采用其逼近非线性系统,用其构成预测模型,将其转化为具有耦合的若干个子系统,利用直接极小化指标函数自适应优化算法辨识其参数,将多变量线性扩张观测器的线性控制输入项改进为关于观测状态和控制输入向量及其微分的向量函数,并由该向量函数的逆向量函数构建当前控制输入向量,因其未知,使用对角回归神经网络逼近控制输入向量函数,采用多变量非线性递推最小二乘法优化对角回归神经网络连接权及多变量线性自抗扰控制参数,综上研究提出在线优化参数的多变量无模型预测神经网络线性自抗扰控制算法。仿真研究表明系统响应精度高,性能好,优于传统的线性自抗扰控制算法。 With regard to control problem of nonlinear system with difficult in model establish,using modified full compact from dynamic linearization with auxiliary variables to approach to the nonlinear system and the system's predicative model was built.The system was transformed into multi-subsystem with coupling.Its parameter estimation was carried out by the adaptive optimization algorithm for direct minimization of index function.Linear control input of a multivariable linear expanded observer was changed into control input vector and vector function of its differential on observation status.Current control input vector was built from the of inverse vector of the function the vector function.Control input vector function was approached by diagonal regression neural network duo to it being unknown.Connection weight of diagonal regression neural network and parameter of the Multivariable model-free predictive neural network linear active disturbance rejection control was optimized by multivariable nonlinearity recursive least squares method.In summary of study above,an algorithm of multivariable model-free predictive neural network active disturbance rejection control with on-line parameter optimization was developed.Simulation result indicates that the algorithm has high quality response,excellent performance and higher against active disturbance rejection.

作者侯小秋 HOU Xiaoqiu(School of Electronics and Controlling Engineering,Heilongjiang University of Science and Technology,Haerbin 150022,China)

机构地区黑龙江科技大学电气与控制工程学院

出处《广西民族大学学报（自然科学版）》 CAS 2023年第4期85-94,共10页 Journal of Guangxi Minzu University ：Natural Science Edition

关键词多变量线性自抗扰控制神经网络控制无模型自适应控制预测控制多变量非线性系统直接极小化指标函数自适应优化算法 Multivariable model free linear neural network active disturbance rejection control Neural network control Model-free adaptive control Predictive control Multivariable nonlinear system Adaptive optimization algorithm for direct minimization of index function

分类号 TP273 [自动化与计算机技术—检测技术与自动化装置]