在线优化参数的神经网络预测监督控制  

Neural network supervisory control with parameter optimization on line

作　　者：侯小秋 HOU Xiaoqiu(School of Electronics and Controlling Engineering,Heilongjiang University of Science and Technology,Harbin 150022,China)

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：《陕西理工大学学报（自然科学版）》2024年第2期38-44,共7页Journal of Shaanxi University of Technology:Natural Science Edition

摘　　要：使用具有辅助变量的全格式动态线性化方法逼近系统,构建了神经网络监督控制预测模型。利用线性跟踪-微分器建立过渡过程,应用线性扩张状态观测器估计输出预测值及其微分,给出了线性PID控制算法。根据对角回归神经网络构成直接逆控制,提出了改进的控制目标函数。依据非线性递推最小二乘法在线优化了PID控制参数和对角回归神经网络的连接权。当系统控制误差大于一定值时,重置PID控制参数。最后提出了在线优化参数的神经网络预测监督控制,克服了已有的神经网络监督控制存在建模难的问题。仿真研究结果表明控制算法的响应具有理想性能。Using the full-state dynamic linearization with auxiliary variables,the nonlinear system is approximated,and a neural network supervised control prediction model was built.The model parameter was estimated by nonlinearity recursive least squares method.Transient process was built by linear tracking differentiator.Using linear expanded observer to estimate output predicative value and its differential,a linear PID control algorithm was obtained.Modified control object function by direct inverse control built from diagonal regression neural network.The parameter of PID control and connect weight of the diagonal regression neural network were optimized on line by nonlinearity recursive least squares method.When error of system control is greater than setting value,parameter of PID control would be resettled.In summary of study above,an algorithm of neural network supervisory control with parameter optimization on line has been developed,which overcomes the problem presented in the already present neural network supervisory.Simulation result indicates that response of the algorithm has excellent performance.

关 键 词：神经网络监督控制 非线性系统 线性PID控制 全格式动态线性化方法 对角回归神经网络 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]