基于前馈神经网络控制的球杆系统位置控制  

Ball and Beam System Position Control Based on Feed Forward Neural Network Control

作　　者：朱坚民[1] 谢平[1] 黄春燕[1] 齐北川 

机构地区：[1]上海理工大学机械工程学院,上海200093

出　　处：《控制工程》2017年第6期1186-1193,共8页Control Engineering of China

基　　金：国家自然科学基金(50975179);上海市教委科研创新(11ZZ136);上海市科委科研计划(13160502500);沪江基金(D14005)资助项目

摘　　要：针对现有球杆系统位置控制精度不高的问题,提出了一种基于前馈神经网络控制的球杆位置控制新方法,该方法控制结构由前馈神经网络控制器、PID控制器以及神经网络辨识器组成。实时控制时首先由神经网络辨识器建立球杆系统的动态逆模型,其更新后的网络参数动态复制给前馈神经网络控制器,再经前馈神经网络控制器运算输出控制量,与PID控制输出叠加后对球杆系统实施控制。以固高GBB1004球杆系统为被控对象进行实验研究。实验结果表明:在反复调整PID控制参数使小球位置控制系统稳定且控制效果较好的前提下,前馈神经网络控制能够进一步提高系统的控制精度,具有更好的动态性能和较高的稳态控制精度。Aiming at the low control accuracy problem of the present ball and beam position control, a new method for ball and beam position control based on feed forward neural network control is proposed in this paper. The feed forward neural network control structure is composed of feed forward neural network controller, PID controller and neural network identifier. During the real time control, the neural network identifier is adopted to establish the dynamic inverse model of the ball and beam system first, and the network parameters are dynamically replicated to the feed forward neural network controller, then through the feed forward neural network controller the output is calculated, which will control the ball and beam system with the addition of the PID controller output. Taking the ball and beam of GBB 1004 as the controlled object, the experimental study is conducted. Experimental results show that： under the condition that the best parameters of PID controller have been chosen for the ball and beam position control system to achieve high control precision, the feed forward neural network control can further improve the steady state accuracy. The feed forward neural network control is proven to have excellent dynamic performance and high steady state control accuracy.

关 键 词：球杆系统 前馈神经网络控制 位置控制 PID控制 实验研究 硬件在环 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]