基于忆阻器的模糊推理系统设计及应用

发布时间：2018-01-08 16:29

本文关键词：基于忆阻器的模糊推理系统设计及应用　出处：《西南大学》2017年硕士论文　论文类型：学位论文

【摘要】：模糊推理系统自提出以来一直被认为是一种最接近人脑计算能力的智能系统,模糊系统具有推理过程容易理解、专家知识利用较好、对样本的要求较低等优点,但它同时又存在人工干预多、推理速度慢、精度较低等缺点,很难实现自适应学习的功能。人工神经网络具有较强的自学习和联想功能,用于模拟人脑的思维功能,且人工干预少,精度较高,但缺点是它不能处理和描述模糊信息,不能很好的利用已有的经验知识,同时它对样本的要求较高。如果将二者有机地结合起来,可起到互补的效果。人脑神经元数量庞大,神经元之间连接复杂,现有的研究大都致力于设计软件计算系统,难以设计出与人脑计算能力相匹配的硬件系统,亟待提出一种可以扩展的简单硬件来模拟人脑单元。纳米级器件忆阻器的提出使类脑硬件电路的实现成为可能。忆阻器是一种无源元件,具有阻值连续可变,非易失性,快速的开关转换特性等优势,在关掉电源后,仍能“记忆”通过的电荷,这与神经元突触的行为类似,可作为神经突触硬件实现的替代物。模糊系统的应用领域在不断扩展,因此研究者们都热衷于找到一种可行的方法来实现一套能够高速实时运行的模糊系统。传统的模糊系统将语言信息在数字系统中进行处理,隶属函数在表示语言变量时转化为二值编码,而随着现代科技对计算速度的要求,就需要计算速度更快的模拟电路形式。取大(max)和取小(min)函数是模糊逻辑里面最重要的部分,如模糊推理就需要取大和取小函数来确定推理结果。本文的主要研究内容包括以下三个部分。(1)为了提供一种硬件实现类脑计算系统的方案,本文提出一种基于忆阻交叉阵列的模糊推理系统,并对复杂函数进行建模来验证设计系统的正确性。首先将传统的前向人工神经网络结构转换成交叉阵列结构,并利用交叉阵列来存储模糊规则,构造了一种新型模糊推理系统,随后提出了一种基于HP忆阻器交叉阵列结构的模糊推理系统硬件电路设计方案。当输入数据后可以根据交叉阵列的输出电压计算得到模糊推理结果,最后用所设计的推理系统对复杂函数建模,数值仿真验证本文设计系统的正确性。(2)利用所提出的模糊推理系统设计了一种基于忆阻交叉阵列的模糊PID控制器,并对典型被控对象进行控制,通过数值仿真得到控制曲面,控制参数变化过程曲线和控制曲线,证明该系统能正确表示各控制曲面,且响应曲线和控制参数都表明该模糊控制系统表现良好。通过与传统PID控制和MATALB模糊PID控制工具进行对比,也表明所设计模糊推理系统在PID控制方面的优势,这也为多变量和多维模糊控制器提供了新的研究思路。(3)基于墨滴扩散的模糊逻辑原理,利用自旋忆阻器的交叉阵列结构存储模糊关系,本文设计了一种基于自旋忆阻器交叉阵列的模糊逻辑门电路,并用LTSPICE电路仿真验证了所设计的模糊门电路的正确性和可行性,此模糊逻辑门电路不仅可以通过更少的步骤实现“与”、“或”、“非”、“异或”等模糊运算,且此模糊逻辑门结构简单,运算速度快,硬件实现成本低、体积小、能耗低、应用范围广,不仅能够用于传统的数字和模拟电路中的各种逻辑门,填补了模糊逻辑门电路的空白,此电路还可以扩展到3值和多值的模糊逻辑。为模糊系统及模糊神经网络硬件实现提供了基础。
[Abstract]:The fuzzy inference system since it has long been considered an intelligent system closest to the computing ability of the human brain, is easy to understand the reasoning process of fuzzy systems, the use of expert knowledge is good, has the advantages of low sample requirement, but it also has more manual intervention, the inference speed slow, low precision, difficult to achieve adaptive the function of learning. The artificial neural network has strong self-learning and associative function, is used to simulate the human brain thinking function, and less manual intervention, higher precision, but the disadvantage is that it cannot describe fuzzy information processing and use the existing knowledge, experience is not very good, while its sample requirements higher. If the two organically, can play a complementary effect. A large number of neurons, neuron complex connection, most of the existing research devoted to the design calculation software system, it is difficult to design The human brain and the ability to match the computing hardware system, to propose a simple hardware can be extended to simulate the human brain cell. Proposed nanoscale devices memristor that realizes brainlike hardware circuit as possible. Memristor is a passive element, with continuous variable resistance, non-volatile, edge switch the conversion characteristics of fast, turn the power off, still can "charge memory" through this, and synaptic actions are similar, can be used as a substitute for hardware implementation. Synaptic fuzzy system used in the field of continuous expansion, so researchers are keen to find a feasible method to achieve a set of can the fuzzy system of high speed. The traditional fuzzy system of language information processing in digital system, the membership functions in the representation language variables into two value encoding, and with modern science and technology on computing speed The request requires faster calculation speed. Taking the form of analog circuit (max) and small (min) function is the most important part of fuzzy logic, such as fuzzy inference requires large and small function to determine the reasoning results. The main contents of this paper include the following three parts. (1 in order to realize the brain like computing system) provides a hardware scheme, this paper proposes a fuzzy inference system based on memristive crossbar array, and the complex functions are modeled to verify the correctness of the design system. Firstly, the traditional feedforward artificial neural network, structural transfer into cross array structure, and using the cross array to store the fuzzy rules, to construct a new fuzzy inference system, and then proposes a fuzzy inference system hardware circuit design of HP memory array structure based on cross resistance. When the input data can be based on cross array The output voltage is calculated by fuzzy reasoning results, finally the inference system design of complex function modeling, numerical simulation to verify the correctness of the design of the system. (2) using fuzzy inference system is proposed to design a fuzzy PID controller based on memristor crossbar array, and the typical controlled object control, control surface through numerical simulation, the control parameter curves and control curve shows that the system can correctly represent the control surfaces, and the response curves and control parameters show that the fuzzy control system. Through good performance with the traditional PID control and fuzzy PID control MATALB tools comparison, also showed that the design of fuzzy inference system in PID control the advantage of this is a multi variable and multi dimension fuzzy controller provides a new research idea. (3) the fuzzy logic principle of droplet diffusion based on the use of spin memristor Cross array memory fuzzy relation, this paper designs a fuzzy logic gate circuit spin memristor based on cross array, and the design of fuzzy circuit correctness and feasibility is proved by the LTSPICE simulation, the fuzzy logic gate circuit can not only through fewer steps to achieve the "and" and "or", "not", "XOR" fuzzy operation, and the fuzzy logic has the advantages of simple structure, fast calculation speed, hardware implementation of low cost, small volume, low energy consumption, wide application range, not only can be used for various types of logic gates the traditional analog and digital circuit, to fill the gaps in the fuzzy logic gate circuit, this circuit can also be extended to more than 3 of the value of fuzzy logic and fuzzy value. The fuzzy neural network system and the hardware implementation provides the basis.

【学位授予单位】：西南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18;TP273

【参考文献】