智能优化方法对神经网络的改进及应用研究
VIP免费
摘 要
伴随着科学技术的飞速发展，人工智能技术越来越受到人们的重视。人工智
能是在生物学、仿生学、控制论、信息论、心理学、语言学、计算机科学等多种
学科互相交叉发展起来的一门综合性的科学。如今人工智能已经在许多的领域中
得到了很好的应用。人工智能学科研究的主要领域包括：机器学习、知识获取、
知识表示、逻辑推理、搜索方法、模式识别、计算机视觉、智能机器人、自动程
序设计等。目前，人工智能的主要技术有模糊理论、人工神经网络、进化计算以
及专家系统。
人工神经网络是一种基于类似人类大脑的推理模型。人工神经网络具有自学
习、并行处理、可以以任何精度逼近非线性函数等的特点，使神经网络在处理非
线性问题时具有一定的优势。BP 神经网络也叫误差逆传播神经网络是在目前应用
的使用比较多的神经网络之一，但是由于它自身结构的问题也存在一些缺点，比
如易陷入极小值、收敛速度慢以及对初始值敏感等。这些缺点阻碍了 BP 神经网络
的应用，基于此本文提出了使用智能优化方法也就是进化计算的方法去改进 BP 神
经网络。智能优化算法一般都是建立在生物智能或物理现象基础上的随机搜索算
法，是一种启发式方法。神经网络的学习本质上是网络权值的优化过程，智能优
化算法去改进 BP 神经网络就是首先寻找一个较优的初始权值。
本文对智能优化方法对神经网络的改进及做了如下的应用研究：
（1）为了提高对空气质量等级预测的准确性，本文提出了一种基于混沌遗传
算法（CGA）改进 BP 神经网络的方法来对空气质量等级进行预测，同时在对预测
的结果处理上也加入了模糊理论的方法。虽然 BP 神经网络算法目前已经应用到预
测、聚类、分类等许多领域，取得了不少的成果，但存在收敛速度慢和易陷入极
小值的缺陷。该改进方法的基本思想是用混沌遗传算法优化 BP 神经网络的初始权
值和阈值。混沌遗传算法结合了混沌运动的遍历性、遗传算法的反演性。将混沌
变量加入遗传算法中，进一步提高了遗传算法的全局搜索能力和收敛速度；将混
沌遗传算法优化后得到的最优解作为 BP 神经网络的初始权值和阈值。利用改进后
的CGA–BP 算法进行空气质量预测，对预测的结果进行模糊化处理。实验结果表
明，该方法对空气质量的预测效果明显好于单纯使用 BP 神经网络的预测结果。
（2）上证指数的预测是一个非常复杂的非线性问题，为了提高对上证指数开
盘指数预测的准确性，本文采用基于混沌粒子群（CPSO）算法对 BP 神经网络算
法改进的方法来进行预测。虽然 BP 神经网络是目前应用最广泛的神经网络，但自
身也有明显的缺点。用混沌粒子群算法改进 BP 神经网络算法的基本思想是利用混
沌粒子群算法来优化 BP 神经网络算法的权值和阈值，在粒子群算法中引入混沌的
元素，用来提高粒子群算法的全局搜索能力。实验结果表明改进后的预测方法，
具有更好的准确性。
（3）对经典的非线性函数拟合问题，使用智能优化算法去改进 BP 神经网络。
在这里选用了遗传算法、粒子群算法和蝙蝠算法三种智能优化方法去改进 BP 神经
网络，通过大量的实验，观察对比单纯使用 BP 神经网络进行拟合的结果。从实验
数据中可以看出，使用混沌遗传算法、混沌粒子群算法以及蝙蝠算法对 BP 神经网
络进行改进后的网络要普遍明显好于只是单纯使用 BP 神经网络进行拟合。而这三
种方法改进的效果和效率也在不同的非线性函数中也各不相同。
关键字：智能优化算法 BP 神经网络 遗传算法 粒子群算法 蝙蝠算法 模糊理论
ABSTRACT
With the rapid development of science and technology, people begin to attach
importance to artificial intelligence. Based on the subjects integration of biology,
bionics, control theory, information theory, psychology and linguistic and computer
science, artificial intelligence becomes a more and more comprehensive subjects.
Nowadays, AI has already successfully applied to a lot of research fields, such as
machine learning, knowledge acquiring, knowledge representation, logical reasoning,
search method, pattern recognition, computer vision, intelligent robot, automatic
programming and etc. Till now, the main technique of AI consists of fuzzy theory,
artificial neural network, evolutionary computing and expert system.
Artificial Neural Network is a reasoning model based on similar human brains.
ANN has the characters of self-learning, paralleled managing and the ability of
impending on non-linear function with any precisions, which makes it have advantages
of dealing with the problems of non-linear functions. BP Neural Network, namely error
back propagation neural network, is one of the most widely used neural network.
However because of the defects of structure, BP Neural Network also has some
disadvantages, for instance, falling into local minimum easily, slow convergence speed
and sensitivity to initial value, which impede applications of BP Neural Network.
Intelligent optimization algorithms and evolutionary computing methods are proposed
to improve BP Neural Network. Intelligent optimization algorithms are random search
heuristic algorithms with the ideas inspired by biological intelligence or physical
phenomena. The essence of neural network is to optimize the network weights.
Intelligent optimization algorithms help improve BP neural network with finding a
better initial weights.
This paper mainly does the following researches about improving BP neural
network with intelligent optimization algorithms:
(1)This paper proposed a method to improve BP neural network, which was based
on chaos genetic algorithm (CGA). This method could be used to improve the accuracy
of air quality forecast. In the meantime, fuzzy theory is used to deal with the results.
Although BP neural network has already applied to the field of forecasting, clustering
and classification, also has some disadvantages, for instance, falling into local minimum
easily, slow convergence speed. CGA considered both the ergodicity of chaotic motion
and the inversion of genetic algorithm (GA). In the process of generating the initial
population, the elements of chaos joined in, the scope of traversal expanded to the entire
scope of variables, which avoid results involved in local optimum and premature
phenomena during the process, and the global searching ability and the convergence
speed are both improved in further. After a generation of initial population, each
individual population contained a network ownership value and threshold value. As the
results, it turned out that applying the CGA-BP algorithm to the air quality forecast and
dealing the result with fuzzy theory had obviously better results than only use the BP
neural network prediction, and it has certain reference function in air quality forecast by
neural network.
(2)The Forecast of the Shanghai Composite Index is a very complex nonlinear
problem. In order to increase the accuracy of Forecast, an improved BP neural network
algorithm is given, based on the chaos and particle swarm optimization algorithm. Till
now, the BP neural network algorithm has been successfully applied to the fields of
forecasting, clustering, classification, etc. But it also has some defects. Our idea for
improving the BP neural network algorithm is to optimize its weights and thresholds by
using the chaos and particle swarm optimization algorithm, that is, add chaos into the
particle swarm optimization algorithm to improve its global search ability. Application
of our treatments to the prediction of the Shanghai Composite Index shows that it is
more accuracy than the original BP neural network algorithm.
(3)As to the classic fitting nonlinear function problem, this paper presents an
improving BP neural network method of using intelligent optimization. Genetic
algorithm, particle swarm optimization and the bat intelligent optimization method are
choosing to apply it. After a lot of experiments, comparing of only using BP neural
network, the improving method has a significantly better result. While the improving
effects in different nonlinear functions also behave unlikely.
Key words: Intelligent optimization, BP neural network, Genetic algorithm,
Particle swarm Optimization algorithm, Bats, Fuzzy theory
目 录
摘 要
ABSTRACT
第一章 绪论 ......................................................... 1
1.1 研究背景及意义 ............................................... 1
1.2 研究现状 ..................................................... 2
1.2.1 智能优化算法研究现状 .................................... 2
1.2.2 人工神经网络研究现状 .................................... 2
1.3 论文研究的主要内容 ........................................... 3
第二章 进化计算 ..................................................... 4
2.1 进化计算概述 ................................................. 4
2.1.1 进化计算的起源 ......................................... 4
2.1.2 进化计算的分支 ......................................... 4
2.1.3 进化计算的优点 ......................................... 6
2.1.4 群智能进化计算 ......................................... 7
2.2 遗传算法 ..................................................... 7
2.2.1 基本概念 ............................................... 7
2.2.2 基本流程 ............................................... 8
2.2.3 控制参数 ............................................... 9
2.2.4 遗传算法的发展 ......................................... 9
2.3 粒子群算法 .................................................. 10
2.3.1 基本原理 .............................................. 10
2.3.2 基本流程 .............................................. 11
2.3.3 控制参数 .............................................. 12
2.3.4 粒子群算法的发展 ...................................... 12
2.4 蝙蝠算法 .................................................... 13
2.4.1 基本概念 .............................................. 13
2.4.2 基本流程 .............................................. 14
2.4.3 参数控制 .............................................. 15
2.4.4 蝙蝠算法的发展 ........................................ 16
第三章 人工神经网络 ................................................ 17
3.1 人工神经网络概述 ............................................ 17
3.1.1 人工神经网络的发展 .................................... 17
3.1.2 神经网络的优点 ........................................ 18
3.1.3 神经元模型 ............................................ 19
3.2 BP 神经网络 ................................................. 24
3.2.1 BP 神经网络的概述 ...................................... 24
3.2.2 BP 神经网络的学习过程 .................................. 25
3.2.3 BP 神经网络的优缺点 .................................... 28
第四章 遗传算法对 BP 神经网络的改进及在空气质量预测的应用 ........... 30
4.1 遗传算法对 BP 神经网络改进的理论基础 ......................... 30
4.1.1 BP 神经网络的不足 ...................................... 30
4.1.2 遗传算法对 BP 神经网络的改进 ........................... 30
4.1.3 模糊数学理论 .......................................... 31
4.2 遗传算法对 BP 神经网络改进的算法流程 ......................... 31
4.2.1 遗传算法部分 .......................................... 31
4.2.2 BP 神经网络部分 ........................................ 33
4.2.3 模糊化处理部分 ........................................ 34
4.2.4 算法的整体流程 ........................................ 34
4.3 在空气质量预测中的应用 ...................................... 35
4.3.1 空气质量问题 .......................................... 35
4.3.2 模型在空气质量中的应用 ................................ 36
4.3.3 空气质量预测实验 ...................................... 37
第五章 粒子群算法对 BP 神经网络的改进及在上证指数预测中的应用 ....... 43
5.1 粒子群算法对 BP 神经网络改进的理论基础 ....................... 43
5.2 粒子群算法对 BP 神经网络改进的算法流程 ....................... 43
5.2.1 粒子群算法部分 ........................................ 43
5.2.2 BP 神经网络部分 ........................................ 45
5.2.3 算法的总体流程 ........................................ 45
5.3 在上证指数预测中的应用 ...................................... 46
5.3.1 上证指数 .............................................. 46
5.3.2 模型在上证指数预测中的应用 ............................ 47
5.3.3 预测实验 .............................................. 47
第六章 智能优化算法对 BP 神经网络的改进及非线性函数拟合中应用 ....... 51
6.1 理论基础 .................................................... 51
相关推荐
-
跨境电商商业计划书模版VIP免费
2025-01-09 6 -
跨境电商方案范文VIP免费
2025-01-09 6 -
创业计划书VIP免费
2025-01-09 6 -
xx生鲜APP计划书VIP免费
2025-01-09 6 -
跨境电商创业园商业计划书(盈利模式)VIP免费
2025-01-09 6 -
跨境电商计划书VIP免费
2025-01-09 7 -
绿色食品电商平台项目计划书VIP免费
2025-01-09 6 -
农产品电子商务商业计划书VIP免费
2025-01-09 6 -
农村电商平台商业计划书VIP免费
2025-01-09 7 -
生鲜商城平台商业计划书VIP免费
2025-01-09 6
作者：侯斌
分类：高等教育资料
价格：15积分
属性：69 页
大小：2.95MB
格式：PDF
时间：2025-01-09
