中山大学硕士学位论文
人工神经网络在证券市场中的应用研究
赵果林
专 业：计算机软件与理论 指导教师：姜云飞 教授 研究方向：人工智能应用
答辩委员会委员(签名)： 主席： 委员：
二○○四年三月
0
0 人工神经网络在证券市场中的应用研究
专 业：计算机软件与理论
硕 士 生：赵果林 指导教师：姜云飞 教授
摘 要
证券市场中成功的交易模式是可以模仿及学习的。证券价格走势实质是一种复杂时序函数。人工神经网络可以通过调节连接权值以任意精度逼近任何连续函数，因此也可以逼近证券价格随时间变换这种函数。选择好的神经网络学习算法，将证券历史数据进行预处理，形成技术指标输入神经网络让它学习成功的交易模式，人工神经网络系统就可以在将来辨别这种模式，产生交易信号。神经网络的学习其实是求最佳的连接权值，好的学习算法可使误差函数尽快收敛。本文详细推导了BP算法，讨论了改进的方法。另外也讨论了输入数据处理，预测方法的选择并提供了基于神经网络算法的证券交易系统软件架构的设计要点并与现有的证券分析软件进行了比较。最后本文用一个人工神经网络壳测试了一支股票，取得了良好的效果。本文得出结论为，人工神经网络的学习及模式识别能力应用于证券市场中可以帮助发现交易机会；选择好的技术指标组合作为神经网络的输入数据，可很大程度的提高人工神经网络交易系统预测的准确度。
关键字：人工神经网络，BP算法，证券市场，预测。
1 人工神经网络在证券市场中的应用研究
Artificial Neural Network Application Study in
Securities Market
Major: Computer Software and Theory Name: Zhao Guo Lin
Supervisor: Professor Jiang Yunfei
Abstract
The successful trading mode in securities market can be learned and copied. The price curve of securities is a complex time-series function. Artificial neural network can approach any continuous function precisely by adjust connection weights, so it can also approach the securities price function which changes along with time. If we choose good neural network learning algorithm and pre-process securities data to form technical indicator as input of neural network to let the network learn successful trading mode, artificial neural network system can produce trading signal by recognize the learned mode in future. The learning of artificial neural network is to find the best connection weights in fact; good learning algorithm can make error function converge faster. In this article, I discussed the backward propagation algorithm and melioration way by step by step reckon. In addition, we also discussed pre-process method of input data and prediction methods. And then provide key point of securities trading system software structure design and explained its strongpoint compare with other securities analysis software. In the end of the paper, I did a test on a stock by use artificial neural network and the result is good. We draw the conclusion that, artificial neural network’s learning and mode recognize ability can help find trading opportunity in securities market and the prediction accuracy of artificial neural network trading system can be greatly improved by choose good technical indicator combination as input data of the neural network.
Keywords： Artificial Neural Network, Backward
Propagation Algorithm, Security Market, Prediction.
2
人工神经网络在证券市场中的应用研究
目 录
摘 要 .................................................................................................................. 1 Abstract .................................................................................................................. 2 目
录 .................................................................................................................. 3
第一章 引 言 ........................................................................................................ 5 第二章 证券市场特性及预测的方法 .......................................................................... 7 第三章．人工神经网络综述 ..................................................................................... 12
1. 人工神经网络(ANN)概述 .................................................................................. 12 2．神经网络的工作原理 ........................................................................................ 14 3．神经网络的特点 ................................................................................................ 15 4．几种典型神经网络简介 .................................................................................... 16 第四章 人工神经网络与证券市场预测的结合 ........................................................ 19
一．人工神经网络应用于证券市场的可行性 ...................................................... 19 二．人工神经网络在证券市场中的应用现状 ...................................................... 20 第五章. BP神经网络算法的改进与网络设计 ......................................................... 22
一、单层感知机神经网络 ...................................................................................... 22 二、多层感知机神经网络概述 .............................................................................. 23 三、BP网络的学习算法――反向传播算法 ......................................................... 24 3.1 3.2 3.3 3.4
BP （Backward-propagation）算法介绍 ............................................... 24 BP算法的数学推导 ................................................................................ 25 BP网络特点总结 .................................................................................... 29 BP算法的改进 ........................................................................................ 30
第六章 神经网络证券预测软件的设计 .................................................................. 32
一 . BP网络的设计 ................................................................................................ 32 二． 网络输入数据的编码及变换 ........................................................................ 33 三. 预测方法 .......................................................................................................... 33 四． 证券市场的预测模型 ..................................................................................... 34 五. 软件架构设计： ............................................................................................. 38
3