自动化学院本科生学位论文格式印刷模2013-5.3

更新时间:2023-11-27 04:25:01 阅读量: 教育文库 文档下载

说明:文章内容仅供预览,部分内容可能不全。下载后的文档,内容与下面显示的完全一致。下载之前请确认下面内容是否您想要的,是否完整无缺。

注:奇偶页页眉相同,居中,宋体,五号。阅后删除此文本框。 兰州交通大学毕业设计(论文) 注:“摘要”中间空两个汉字。“目录、结论和致谢”要求相同。阅后删除此文本框。 注:本部分摘要内容由学生和指导教 师按照后面规范确定,这里仅仅是段落格式要求。阅后删除此文本框。 摘 要 基于神经网络的模型参考自适应控制既具有神经网络的自适应和自学习能力,又提高了控制的鲁棒性和实时性,因此得到了广泛应用并取得了不少研究成果。但由于传统神经网络控制器采用BP学习算法,存在着收敛速度慢,易陷入局部极小等问题,影响了该模型的进一步应用。基于此,本文对传统BP学习算法进行改进,并将此应用到模型参考自适应控制神经网络控制器的设计中,和数字锁相环控制组成复合控制,对感应加热电源进行控制,仿真结果表明该算法是十分有效的。主要研究内容如下:

(1) 提出了改进的分层递阶BP(HBP)和改进的双向权值调整两种学习算法。HBP网络结构被划分为多个单独的BP网络,每个BP网络分别用单独的学习速率和误差来训练权值。通过应用HBP算法在非线性系统辨识中进行仿真,证明了该方法的有效性和可行性。改进的双向权值调整学习算法是单隐层神经网络结构,将神经网络权值的调整分成两个阶段:正向阶段调整隐含层到输出层的权值,反向阶段调整输入层到隐含层的权值。这样构建的网络结构简单,弥补了传统BP算法的不足。

(2) 基于改进的HBP算法,提出了应用HBP算法训练RBF网络连接权的改进RBF神经网络,仿真结果表明该算法能有效提高了RBF网络的性能和训练速率。

(3) 通过采用改进双向权值调整学习的BP学习算法完成对被控对象的辨识,改进RBF神经网络训练控制器参数,学习速率自调整的BP算法训练自适应滤波器,提出改进的神经网络模型参考自适应控制器。通过对该系统仿真,结果表明该控制方法是可行的和有效的。

(4) 将改进的神经网络模型参考自适应控制器和数字锁相环控制相结合,提出复合控制器,并对感应加热电源进行控制。当逆变器开关频率与负载固有谐振频率误差值大于或等于偏差设定值时,采用单纯的神经网络模型参考自适应控制进行频率修正;当频率误差值小于偏差设定值时,采用数字锁相环控制,进行频率和相位同时高精度修正。与传统锁相环控制感应加热电源进行对比,复合控制提高了系统的快速性,改进了感应加热电源的性能。

关键词:模型参考自适应控制;神经网络;BP算法;RBF算法;锁相环

- I -

注:次页为说明页,阅后删除此页和此文本框。 兰州交通大学毕业设计(论文) 中文摘要部分格式要求: (1)“摘要”是摘要部分的标题,不可省略。

标题“摘要”选用模板中的样式所定义的“标题1”,再居中;或者手动设置成字体:黑体,居中,字号:小三,1.5倍行距,段后11磅,段前为0。

(2) 论文摘要应不少于400字。

(3) 摘要正文选用模板中的样式所定义的“正文”,每段落首行缩进2个汉字;或者手动设置成每段落首行缩进2个汉字,字体:宋体,字号:小四,行距:多倍行距 1.25,间距:段前、段后均为0行,取消网格对齐选项。

(4) 摘要正文后,列出3-5个关键词。“关键词:”是关键词部分的引导,不可省略。关键词请尽量用《汉语主题词表》等词表提供的规范词。

(5) 关键词与摘要之间空一行,多倍行距1.25。关键词词间用分号间隔,末尾不加标点,3-5个,黑体,小四,加粗。

中文摘要部分格式注意事项:

(1) 论文摘要是学位论文的缩影,文字要简练、明确。内容要包括目的、方法、结果和结论,应突出论文的新见解部分。单位制一律换算成国际标准计量单位制,除特别情况外,数字一律用阿拉伯数码。文中不允许出现插图。重要的表格可以写入。 (2) 摘要中的编号和标点用Times New Roman字体,摘要和结论中如需要出现序号的采用“(1)、(2)”的形式,字体为Times New Roman(英文中编号与此要求相同)。

- II -

兰州交通大学毕业设计(论文)

Abstract

Neural Networks model reference adaptive control (NNMRAC) not only possesses the neural networks’ merits of adaptive and self learning ability, but also improves the control system’s performance in robust and real-time. NNMRAC has attracted so many researchers’ interest and attention, and lots of research results have been achieved. However, traditional NNMRAC adopts BP learning algorithm, which limits its further extended applications due to slow convergence speed and easily falling into local minimum of BP algorithm. Based on it, the thesis implements improvements on traditional BP learning algorithm, and adopts the improved BP learning algorithm to NNMRAC to construct the combination control on the induction heating power with digital phase-locked loop (DPLL), together. The simulation results indicate that the proposed algorithm is quite effective. The main contents are below.

(1) To propose the two improved BP learning algorithm called as the improved hierarchical back propagation (HBP) algorithm and the improved Bi-phases weights adjusting learning algorithm, respectfully. The HBP neural networks can be divided into several individual BP neural networks, the weights in each network are trained by individual error cost and learning rate. Simulations in the nonlinear system identification show the effectiveness and feasibility of the proposed HBP algorithm. The improved bi-phases weights adjusting learning algorithm has a single hidden layer structure, and the weights are adjusted in two stages. The weights from hidden layer to output layer are adjusted at forward process, and the weights from input layer to hidden layer are adjusted at backward process. Thus the networks’ structure is simplified and the shortcomings of the traditional BP algorithm are remedied.

(2) To propose an improved RBF neural networks with HBP based. In this algorithm the connection weights of RBF networks are trained by HBP algorithm. The simulation results show that the algorithm improves the performance of the RBF networks and the training rate.

(3) To propose the improved NNMRAC applying the improved Bi-phases weights adjusting learning algorithm to identify the controlled object, and the improved RBF neural networks to train the parameters of controller, and the learning rate self-tuning BP algorithm to learn the adaptive filter. The simulations are carried out to validate the effectiveness and feasibility of the system.

(4) To propose the combination controller based on the former mentioned NNMRAC and DPLL to control the induction heating power control system. When the error of inverter switching frequency and the load inherent resonant frequency is greater than or equal to the bias setting value, NNMRAC control the frequency alone. When the frequency error is less than bias setting value, DPLL controller is implemented to modify the frequency and phase

- III -

兰州交通大学毕业设计(论文)

with high-precision at the same time. Comparing with the traditional DPLL induction heating power control system, the combination controller improves the system’s response speed and accordingly improves the performance of the induction heating power control system.

Key Words: Model reference adaptive control, Neural networks, Back propagation algorithm, Radial basis function networks, Phase-lock loop

注:英文内容标点符号后需要有一个空格,阅后删除此页和此文本框。

- IV -

注:次页为说明页,阅后删除此页和此文本框。 兰州交通大学毕业设计(论文) 英文摘要部分格式要求:

(1) “英文摘要”内容应与“中文摘要”对应。使用第三人称,最好采用现在时态和被动语态编写。

(2) “Abstract”不可省略。标题“Abstract”选用模板中的样式所定义的“标题1”,再居中;或者手动设置成字体:黑体,居中,字号:小三,多倍行距1.5倍行距,段后11磅,段前为0。

(3) Abstract正文选用设置成每段落首行缩进2字,字体:Times New Roman,字号:小四,行距:多倍行距 1.25,间距:段前、段后均为0行,取消网格对齐选项。

(4) Key Words与Abstract之间空一行。Key Words与中文“关键词”一致。词间用逗号间隔,末尾不加标点,3-5个,Times New Roman,小四,加粗。

英文摘要部分格式注意事项: (1) 不允许使用机器翻译。

(2) 英文关键词之间用逗号隔开且换行后不缩进。英文摘要关键词中每一个关键词 的第一个单词首字母大写。英文摘要中的标点符号均要求用英文中新罗马下的标点符号。

(3) 中英文格式和内容均应一致。

- V -

本文来源:https://www.bwwdw.com/article/xjqt.html

Top