【正文】 為我們提供的良好的做畢業(yè)設(shè)計(jì)的環(huán)境，這一切都是我們?nèi)〉贸删偷那疤?。感恩大學(xué)四年，感謝伴著我們成長(zhǎng)的老師們，三個(gè)月的畢業(yè)設(shè)計(jì)讓我真正完全的投入到學(xué)習(xí)中，系統(tǒng)的溫故了原有的知識(shí)體系，更加深化的明白了每一門專業(yè)課程都是如何去使用的，真正的從理論走向了實(shí)踐，一直在摸索中不斷前進(jìn)，每一步的成功都會(huì)帶來(lái)喜悅，到最后系統(tǒng)的將整個(gè)程序做出來(lái)，喜不自禁，這其中我想說(shuō)的是，不要認(rèn)定它是難的，只要肯努力肯去做，過(guò)程雖苦卻苦中有樂(lè)！最后再一次感謝所有在畢業(yè)設(shè)計(jì)中曾經(jīng)幫助過(guò)我的良師益友，以及在設(shè)計(jì)中被我引用或參考的文獻(xiàn)作者，謝謝你們！　　參考文獻(xiàn)[1] :電子工業(yè)出版[2] 周開(kāi)利，:清華大學(xué)出版社[3] [4] :重慶大學(xué)出版社 [5] :高等教育出版社[6] :科學(xué)出版社[7] 葛哲學(xué)， :電子工業(yè)出版[8] 馮金剛，:工業(yè)學(xué)院學(xué)報(bào)[9] 蔣宗禮 :高等教育出版社[10] 聞新，周露，王丹力，:科學(xué)出版社[11] :電子科技大學(xué)出版社,[12] 聞新，周露，李翔，:科學(xué)出版社[13] 王繼龍 [14] 0thmar Kyas夏俊杰，:[15] 單紀(jì)文， 附 錄附錄A　英文原文Neural Networks and Genetic AlgorithmApproaches to AutoDesign of Fuzzy SystemsHideyuki TAKAGI and Michael LEEComputer Science Division, University of California, Berkeley, CA 94720takagi@, lce@, FAX (510)6425775Abstract: This paper presents Neural Network and Genetic Algorithm approaches to fuzzy system design, which aims to shorten development time and increase system performance. An approach that uses neural network to represent multidimensional nonlinear membership functions and an approach to tune membership function parameters are given. A genetic algorithm approach that integrates and automates three fuzzy system design stages is also proposed.1 Introduction Fuzzy system is often manual design. This caused the two problems, one is due to the manual design is time consuming, so the development cost is very high, there is no guarantee that get the best solution. In order to shorten the development time and improve the performance of fuzzy system, there are two independent approaches, development support tools and automatic design method. The former includes auxiliary fuzzy system design development environment. Many environmental have mercial use. The latter introduces the design of automatic technology. Although automated design cannot ensure to get the optimal solution, they are still desirable manual skills, because design is a guide to and in accordance with certain criteria of the optimal solution.. There are three major design decisions to make when designing fuzzy systems:(1) determine the number of fuzzy rules,(2) determine the shape of membership functions.(3) changes in the parametersIn addition, two other decision must be made:(4) determine the number of input variables(5) determine the reasoning methods (1) and (2) to coordinate with each other to determine how to override the input space. A high degree of mutual dependence between them. (3) is used to determine the TSK (Takagi Sugeno Kang) model of the coefficient of linear equation in [1], or to determine the membership function and part of the Mamdani model [2]. (4) in accordance with all relevant decided to a minimum set of input variables, calculation of the required target decision or control values. As reverse elimination (4) and standards of information technology is often used in this design. (5) is equivalent to which one decides to use fuzzy operator and fuzzy method. Although by several algorithms and fuzzy reasoning methods have been proposed, still don39。t have to choose their standard . Show that dynamic reasoning methods, according to this reasoning environment results in the performance and fault tolerance is higher than any fixed method of reasoning.Neural network model (in the more general gradient) and neural network based on genetic algorithm (the most mon gradient based) and genetic algorithm is used for automatic design of the fuzzy system. The method based on neural network is mainly used to design the fuzzy membership functions. There are two main methods:(a) direct the multidimensional fuzzy membership function of the design:The method firstly determined by rules. The number of the database And then through the level of training to determine the membership function of each cluster. More details will be presented in chapter 2.(b) indirect multidimensional fuzzy membership function of the design:This approach by bining a onedimensional fuzzy membership function to construct multidimensional fuzzy membership functions. Membership function gradient technique is used to adjust try to reduce the expectations of fuzzy system output and the actual production of the required output error.The virtue of the first method is that it can directly produce nonlinear multidimensional fuzzy membership functions。 There is no need to through the bination of one dimension fuzzy membership function to construct multidimensional fuzzy membership functions. The second method has an advantage by monitoring last performance to adjust the fuzzy system. Both methods will be in the second chapter.Many methods based on genetic algorithm and the method 2 in essentially the same: one dimensional membership function by using the genetic algorithm in the form of automatic adjustment. Many of these methods consider only one or two design problems mentioned earlier. In the third chapter, we introduce a method of three design issues to be considered at the same time.2 neural network method multidimensional fuzzy partition of input space directlyThe method using neural network to realize the multidimensional nonlinear membership functions, called fuzzy reasoning based on NN.The advantage of this method is that it can produce nonlinear multidimensional fuzzy membership functions. , used in traditional fuzzy system for the early part of the one dimensional membership function is independent design, and then bine the indirect achieve multidimensional fuzzy membership functions. Can say, neural network method in bined with the operation of absorption by the neural network is one of the traditional fuzzy system more general form. When the input variables are independent of the indirect design method of the traditional have a problem. For example, design a fuzzy system based on temperature and humidity as the input of the air conditioning control system. In traditional design method of fuzz