【作者】 邱焕耀；

【导师】 毛宗源；

【作者基本信息】 华南理工大学 ， 控制理论与控制工程， 1999， 博士

【摘要】 智能控制中模糊控制和人工神经网络是较为成功的两个研究课题，模糊控制和人工神经网络的交叉结合的研究成了自动控制研究中的热点之一，并在模糊神经网络方面取得了一批有意义的研究成果。模糊神经网络的更深层的理论和应用研究对于促进智能控制在生产实践中的应用，提高劳动生产率具有十分重要的意义。 本文首先在总结国内外模糊控制和神经网络研究的基础上，提出将两者交叉结合得到模糊神经网络，讨论了模糊神经网络的多种类型的应用途径，分析了模糊神经网络的工作机理。它既能充分发挥两者优势互补的特点，又能有效地克服两者的缺点。 本文在分析和指出了模糊神经网络存在着网络性能不佳，学习效率不理想，网络难以选择最佳结构，甚至会陷入局部极值等问题，提出了采用结构学习模糊神经网络来改进网络性能的办法，详细论述了结构学习模糊神经网络控制的结构和学习算法，并构成了结构学习模糊神经网络控制系统，对三层模糊神经网络和多层模糊神经网络结构学习算法进行了研究。 本文对结构学习模糊神经网络的工作机理进行了阐述，与人脑的思维比较，提出了结构学习模糊神经网络的控制策略和学习规律，从粗到精的学习方法是结构学习模糊神经网络学习过程的本质。结构学习模糊神经网络的优点是可以进行网络结构的学习，得到神经网络的最佳结构，克服了单凭设计经验来选择网络结构的随意性，提高了模糊神经网络的学习收敛速度。 本文论述了感应电动机的解耦控制原理，在此基础上提出了解耦变结构控制，并对变结构引起的系统抖振问题进行了研究，提出了抑制和消除系统抖振，提高控制性能的方法。本文还讨论了采用模糊控制的感应电动机解耦控制系统，证实了模糊控制的引入抑制了系统抖振的影响。模糊滑模控制引入后会存在静态误差，消除静态误差的有效方法是修改控制规律。最后，还讨论了结构学习模糊神经网络在感应电动机解耦控制系统中的应用。 本文讨论了结构学习模糊神经网络在GPS全球卫星定位系统中的应用，讨论了GPS卫星定位系统中园范围呼叫方法存在的问题，更多还原

【Abstract】 Fuzzy system and artificial neural network are two of the successful directions in intelligent control. The research of intersection and integration of fuzzy control and artificial network has become one of the popular areas in automation, and a lot of significant achievement is got in fuzzy neural network (FNN). The embedded theory and application research in FNN is much significant for promting the application of intelligent control in productive practice and increasing the rate of production.On the basis of summarization of fuzzy control and neural network in domestic and abroad, their intersection and integration gives rise to FNN is proposed in the paper. Many kinds of FNN and their application are discussed. The mechanism of FNN is analyzed. It can not only bring into full play the superiority complment of fuzzy control and neural network but also overcome their shortcomeing.There still exists difficulty in FNN, such as the network performace is not so good, the efficiency of learning is not ideal, it is difficult to select optimal network structure, and even it may fall into local extreme value. Fuzzy neural network with structure learning (SL_FNN) is applying to promote the performance of network. The structure and learning algorithm of SLFNN is fully discussed, and SLFNN control system is obtained. The structure learning algorithm of three-layer and mutiple-layer SL_FNN is proposed.The mechanism of SLFNN is explained in the paper. The comparison with the brain of human being gives rise the control strategy and learning principle of SLFNN. The learning method from rough to fine is the essential characteristic of the learning process of SL_FNN. The superiority of SLFNN is that nerwork structure learning is achieved, and optimal one is reached. It overcomes the uncertainty of selecting network structure just by design experiment, and increases the FNN learning convergence velocity.更多还原