【作者】 袁小芳；

【导师】 王耀南；

【作者基本信息】 湖南大学 ， 控制理论与控制工程， 2006， 硕士

【摘要】 近年来,模糊逻辑、神经网络、进化算法、混沌优化、支持向量机等计算智能理论和方法是国内外电子工程、自动化、计算机科学等领域研究的热门前沿课题之一,取得了很大的发展,尤其是在控制领域得到了深入研究和应用。本论文以混沌优化、支持向量机二种计算智能方法为主,以此研究非线性系统的建模与控制。论文首先介绍了混沌理论的相关知识,接着详细地描述了混沌优化算法,在分析混沌优化算法特点的基础上,提出了一种并行混沌优化算法融合单纯形法的优化算法,接着分析了该优化算法的优化性能。随后,论文将该优化算法应用于多种类型模型的系统辨识,既有线性系统,又有非线性系统,取得了较好的仿真结果。支持向量机方法是一种新的机器学习算法,其原理是建立在统计学习理论的VC维理论和结构风险最小原理基础上,表现出了许多优越于人工神经网络的优点,如:全局最优、学习训练速度快、范化能力强等等。论文描述了支持向量机的基本知识,从支持向量机分类和回归两个方面介绍了其基本原理和理论,描述了核函数的形式和网络结构,并着重介绍了支持向量机的学习训练算法。其后,针对非线性系统的模型辨识及其逆模型辨识等建模问题,论文考虑将非线性逼近性能强、学习能力强的支持向量机应用于此类建模问题,以此作为控制器设计的基础。文章描述了基于支持向量机的建模步骤和具体实现方法,并且以仿真的形式验证了其实际性能。接着,论文研究了基于支持向量机逆模型的控制器设计,提出了支持向量机直接逆模型控制、PID补偿的支持向量机逆模型控制,通过仿真研究验证了所研究的这二种逆模型控制器的性能和效果;考虑到支持向量机具有很强的学习能力,文章还研究了利用支持向量机去提高模糊推理系统的学习能力、优化能力,从而提出了一种支持向量机-模糊推理自学习控制器,并对比研究了梯度法、混沌优化算法这二种学习算法,并给出了相应的仿真结果。更多还原

【Abstract】 Recently, computational intelligence theory and method, such as fuzzy logic, artificial neural networks, genetic algorithms, chaotic optimization algorithms (COA) and support vector machines (SVM), has been a hot research area in electronic engineering, automation, computer science both outside and inside our country. At the mean time, there are many great progresses in the research of computational intelligence, especially in the control system design. This paper focuses on two kinds of computational intelligence, that is, COA and SVM, for modelling and controller design of nonlinear systems.Firstly, this paper introduces the relative knowledge about chaos, and describes COA in detail. For improving the search ability and convergence of COA, this paper proposes a improved search algorithm which is the combination of parallel COA and simplex method. Simplex method has good local search ability, thus it is used for improving the local search ability for parallel COA in this proposed algorithms. The capability of the proposed algorithms is analysed and it is applied in system identification with diverse types.Thereafter, this paper introduces basic knowledge about SVM, another computational intelligence method, in the views of classification and regression. Then the main learning algorithm of SVM is presented in detail.As SVM have good ability for nonlinear system approximation, in this paper, it is employed for system identification as well as its inverse model, this is the base for controller design.In succession, this paper focuses on the design of inverse model control system using SVM, and compares three kinds of inverse model control based on inverse model, that is, direct inverse control, PID compensated inverse model control. The control capability of these two controllers are simulated which validate the performance of controllers.As fuzzy inference system(FIS) has poor self-learning ability, for the improving the learning ability of FIS, this paper presents a SVM-FIS self-learning controller. Both grads descending algorithm and chaotic optimization for the training of SVM-FIS self-learning controller are presented. Simulations show that the proposed controller has good performance.更多还原