【作者】 曾祥光；

【导师】 杨名利；

【作者基本信息】 西南交通大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 倒立摆是一个典型的单输入多输出、非线性、高阶次的不稳定系统,研究倒立摆的控制不仅能反映控制理论中有关非线性、鲁棒性以及跟踪问题等许多关键问题,同时对工业复杂对象的控制也有着重要的应用价值。微粒群优化算法(Particle Swarm Optimization algorithm,PSO)是近年来提出的一种新型的基于群体智能的进化算法,它具有算法简单、收敛速度较快,所需领域知识少的特点。本文在对倒立摆、PSO算法研究现状进行综述的基础上,进行了基于PSO算法优化设计控制系统的研究,论文主要工作包括:(1)基于牛顿力学原理建立了一级直线倒立摆数学模型,介绍了倒立摆LQR(Linear Quadratic Regulator)最优控制方法,并利用MATLAB7研究了倒立摆LQR控制性能。(2)将微粒群算法应用于传统PID控制器参数优化整定,通过对不同对象的控制系统仿真实验结果表明,与传统PID控制器整定方法相比,控制系统具有更佳的闭环控制性能。(3)为克服BP(Back-Propagaion)算法不足,本文研究利用PSO算法作为多层前馈神经网络训练算法以实现非线性函数逼近及模式识别等,通过对不同非线性函数的辨识及模式识别实验结果表明,PSO算法作为神经网络训练算法是可行的。为提高逼近精度,采用“种群爆炸”思想对PSO算法进行改进,实验结果表明,改进方法是有效的。为得到全局最优的BP网络训练算法,文中还将PSO算法与BP算法结合进行网络训练开展了研究工作,实验表明,该方法能克服BP算法不足,提高网络训练速度和精度。(4)针对一级直线倒立摆这一复杂非线性对象,本文提出了一种基于PSO算法训练BP网络连接权值与阈值的神经网络控制方法以实现倒立摆控制,仿真结果表明了该方法的良好性能。文章最后对全文的工作进行总结,并且提出了进一步研究的方向。更多还原

【Abstract】 The inverted pendulum system is a typical single input and multiple outputs,nonlinear,high order,natural unstable system. Research on the accurate control of the inverted pendulum not only reflects several joints-in control theory,such as,nonlinear problems,robustness,as well as tracking,but also has great engineering value to the complex industrial objects.Recently,Particle Swarm Optimization(PSO)algorithm comes forth as another intelligent algorithm.It is simple with concept, parameters and implementation.Inverted pendulum、PSO and its researchment actuality are summarized firstly,then PSO is applied to optimize and design the control systems.The main contributions given in this dissertation are as follows:(1)Using Newton’s mechanicstheory to establish the linear level inverted pendulum mathematical model.LQR(Linear Quadratic Regulator)controller in modern control theory is designed,and simulation performances are given by MATLAB7.0.(2)PSO is proposed to optimize the parameters of the conventional PID controller.The simulation results of the different control systems show that the optimal PIP controller based on PSO has a satisfying performance and is better than the conventional PID controller based on the conventional.(3)PSO algorithm is used to train the weights and the thresholds of Multilayer Feedforward Neural Networks(MFNN)instead of Back Propagation(BP)algorithm.The NN trained by PSO is applied to identify non-linear function and pattern recognition.The experimental results show that the proposed method is effective but low-precision than BP algorithm.To improve the global searching capability of PSO,the concept of ’Particle Swarm Exploding’ is introduced into the PSO,experimental results show that the proposed method is effective.The hybrid algorithm combining PSO algorithm with BP algorithm is used to train the MFNN.Defects of conventional BP algorithm,i.e.the slow convergence of weight and threshold learning,premature result,and the slow training speed of PSO,are settled by it.(4)A NNC(Neural Network Controller)is made based on PSO algorithm to control inverted pendulum and theexperiment results prove the good effectiveness of NNC method based on PSO.Finally some conclusions and future researches are drawn in this dissertation.更多还原