多层前向型人工神经网络的设计及算法研究+MATLAB程序
时间:2020-05-30 16:28 来源:毕业论文 作者:毕业论文 点击:次
| 摘要人工神经网络是一门交叉性学科，它不仅融合了诸如数学、计算机科学、神经学、控制信息科学等多个学科的知识，而且它也广泛地应用到众多领域。目前，神经网络的众多模型中，应用最广的是前向型神经网络，而关于人工神经网络的研究也多集中在前向型网络上，其中基于误差反向传播算法的网络成为研究热点。不过，这些研究多以单隐层的网络结构为主，鲜少考虑多隐层结构。本论文基于误差反向传播算法，对多层前向神经网络进行了研究。 基于误差反向传播算法，推导出三层前向神经网络的权值迭代公式，并用非线性曲线拟合实验证明该公式是有效的。同时对三层前向网络进行了性能分析，通过与双层前向网络的曲线拟合实验的比较，证明三层前向网在函数拟合，尤其是周期函数拟合上是具有优势的。50098 毕业论文关键词 神经网路 误差反向传播算法 函数拟合 Title Design of Multilayer Forward Artificial Neural Network andIts Algorithm Abstract As a cross discipline, artificial neural network contains knowledge ofmathematics, computer science, neuroscience as well as control information scienceand it is extensively used in many fields. At present, forward neural network ismost widely used among various neural network models and the research of artificialneural network mostly focus on it. Network based on error back propagationalgorithm becomes a research hot issue. However, single hidden layer structureis heavily used while multilayer hidden network structure rarely in theseresearches. This paper will study multi-layer forward neural network based on errorback propagation algorithm, deduce weight iteration formula of three layer forwardneural network and prove its effectiveness with nonlinear curve fittingexperiment. At the same time this paper will analyze the performance of three layerforward neural network and compare it with the curve fitting experiment of doublelayer forward network to prove that three layer forward neural network has anadvantage over function fitting especially over periodic function fitting. Keywords Neural network error back propagation algorithm function fitting 目次 1绪论1 1.1研究背景及意义.1 1.2研究现状.1 1.3研究内容.3 2理论基础4 2.1人工神经网络概述.4 2.2多层前向型神经网络.6 2.3误差反向传播算法简介.8 3三层前向型人工神经网络.9 3.1权值迭代公式的推导10 3.2学习算法流程和步骤.13 3.3有效性分析.14 4性能分析16 4.1优点与问题16 4.2网络优化和算法改进18 4.3与双层前向型神经网络的比较20 结论24 致谢25 参考文献26
附录A三层前向网MATLAB实现程序.28 |