自适应控制在随动系统测试装置中的应用与研究

【作者】 王跃武；

【导师】 侯远龙；

【作者基本信息】 南京理工大学 ， 机械制造及其自动化， 2009， 硕士

【摘要】 关于人工神经网络与自适应结合的研究,近年来已经成为智能控制学科的热点之一。由于自适应控制具有很强的鲁棒性,神经网络则具有自学习功能和良好的容错能力,研究如何把神经网络与自适应控制结合起来,发挥各自的优势,对控制理论与应用研究具有重要的意义。本文针对随动加载与测试系统,主要对其中的阻力矩加载系统建模与控制等方面进行研究。本文首先阐述了神经网络的基本问题和理论,详细研究了典型的前向神经网络(BP网络)的学习和训练算法,针对BP网络训练算法速度较慢且易于陷入局部极小点的不足,提出LM—BP算法。接着本文研究了神经网络的模型辨识问题,分析了神经网络辨识的基本结构,讨论了BP网络辨识问题。并采集了一组磁粉制动器恒速下的实验数据,进行了基于神经网络的非线性系统离线的正模型辨识,并分析了辨识结果。最后,本文采用神经网络与自适应相结合的方法构造了基于辨识模型的智能控制器。接着结合阻力矩加载系统的控制研究了神经网络间接自校正控制器算法,并进行了仿真研究。经过大量的系统仿真试验,所设计的间接自校正控制器可以使系统具有良好的动、静态性能,能够实现对阻力矩加载系统精确的控制。更多还原

【Abstract】 The research of combination of neural network and adaptive theory has been an important topic in the intelligent control. As neural network adaptive control not only has the good robustness as that in the adaptive systems, but also has the ability of self-learning and good fault-tolerant, it is very interesting for the control theory and application to research how to combine neural network with adaptive control.The paper mainly researches the modeling and control for the resisting moment loading system in a loading and test device for servo system. The paper firstly expounds the basic problem and theory of neural network, and a typical multi-layer feed-forward artificial neural networks named BP network has been studied. But traditional BP neural network has many defects, such as slow training velocity and converge to a local minimum point, while LM-BP algorithm has much better performance.The paper then researches the model identification based on neural network, presents the normal structures of neural network identification and discusses the identification for BP neural network. According to a set of experiment data of magnetic particle brake running in constant speed, the paper has made system identifications of positive and offline model based on neural networks, and analyses identification result.Finally, the method incorporates adaptive theory with neural network to obtain an intelligent controller, on account of model identification. The paper, combining the method and the control of resisting moment loading system, researches the neural network of indirect self-tuning controller algorithm, and carries out simulation studies. A set of simulation results shows that the design of indirect self-tuning controller can make the system has good dynamic and static performance, and realizes precise control of resisting moment loading system.更多还原