基于神经网络的重力卸载系统控制器设计  

Design of controller for gravity unloading system based on neural network

作　　者：孙一斌 王丽梅[1] SUN Yibin;WANG Limei(School of Electrical Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]沈阳工业大学电气工程学院,沈阳110870

出　　处：《中国空间科学技术》2021年第1期113-119,共7页Chinese Space Science and Technology

摘　　要：吊挂系统是地面模拟空间机械臂重力卸载试验的重要方法之一。针对传统PID控制方式动作响应慢、鲁棒性差等缺点,提出了一种基于径向基函数(RBF)神经网络的智能控制方式。该方式有很强的非线性拟合能力,且学习规则简单,可映射任意复杂的非线性关系,便于计算机实现。利用该特性,设计了一种重力卸载精度较PID控制方式更高的控制器。该控制器应用正交最小二乘法辨识的控制模型,在带负荷的伺服电机数学模型条件下,使用基于梯度下降法更新权重的RBF神经网络,最后通过s-函数编写,得到Matlab仿真图像。仿真结果与PID控制方法相比响应迅速,鲁棒性强,重力卸载精度更高(98%)。The hanging system is one of the important methods to simulate the gravity unloading experiment of a space manipulator on the ground.To overcome the shortcomings of slow response and poor robustness of traditional PID control modes,an intelligent control method based on radial basis function(RBF)neural network was presented.This method has strong non-linear fitting ability and simple learning rules.It can map any complex nonlinear relationship,and is convenient for computer implementation.Taking advantage of this feature,a controller with higher precision of gravity unloading than PID control was designed.The control model identified by the controller using orthogonal least squares method,updates the weight RBF neural network by using a gradient descent method under the mathematical model of the servo motor with load.Finally,the simulation image of Matlab was obtained by writing s-function.Compared with PID control method,the simulation results have fast response,strong robustness and higher accuracy of gravity unloading(98%).

关 键 词：吊挂 重力卸载 梯度下降法 最小二乘法 RBF神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]