遗传算法及BP神经网络融合策略在股市预测中的应用
|
摘要
随着经济的发展和人们投资意识的转变,股市已成为现代人生活中的一个重要组成部分,股市投资已成为社会公众谈论的中心之一,而股市的健康发展和繁荣也成为管理者和投资者关心和研究的重点。股市投资的收益与风险往往是成正比的,即投资收益越高,可能冒的风险越大。因此,股市预测方法的研究具有极其重要的应用价值和理论意义。但是股价系统内部结构的复杂性、外部因素的多变性决定了这项任务的艰巨性,而传统的预测工具已不能满足这种需要。
本文在深入分析股市投资理论和股价预测方法的基础上,使用BP神经网络作为股市预测的网络模型。股市市场是一个极其复杂的非线性动力学系统,而神经网络具有很强的非线性逼近能力和自学习、白适应等特性,实验证明,利用神经网络对股市建模可以取得较好的预测效果。因为股市市场的走势看起来杂乱无章,但实际上有其内在的变化规律,而这正是神经网络预测股市的基础。BP网络通过对以往历史数据的学习,找出股市运行的内在规律,并将其存储在网络具体的权值、阈值中,用以预测未来的走势,尤其对于短期的预测效果更为明显。
然而,BP神经网络存在学习收敛速度缓慢、易陷入局部极小点等缺点,使其对股市预测的效果不能令人满意。鉴于此,本文采用遗传算法与BP神经网络的融合策略,来达到克服其缺点的目的。
遗传算法与BP神经网络相结合运用于股市预测的研究早已不是什么新鲜事了,但对于换手率作为因子能否用于股市预测?如何在预测网络模型中加入换手率因子?到目前为止还没有学者作过研究。作为首次研究,本文提出了两种将换手率作为因子加入到预测网络模型的方案:一,换手率和收盘价同时作为网络的输入数据;二,平均换手率作为隐含层阈值的附加值。对每个方案开发一个预测系统,通过系统的实际运行来判断换手率作为因子能否用于股市预测以及预测效果如何,从而达到研究遗传算法与BP神经网络融合策略在时间序列预测中应用的目的。
With the economic growth and the conversion of people's investment consciousness, stock has become an important part of people's life in modern time. Investment in stock has greatly become one of focuses of public topic. How to keep the development and boost stock market is becoming the emphasis of concern and research of manager and investor. The proceeds of stock investment always equal the risk. That means the good proceeds is based on the high risk of failure. Therefore, the study of stock prediction method has great application value and theoretical significance. The complexity of inside structure and levity of exterior complication in system of stock market make stock market predication a complex problem. The traditional methods and tools have not met its challenge, the thesis presents a method of modeling stock market using neural network based on thorough study of stock investment theories and stock prediction methods.
On base of analysis of investment theory and the stock market price prediction method, BP neural network was utilized as a network model for stock market prediction. The stock market is an extremely complex non-linear dynamical system, and neural network has strong nonlinear approximation ability and self-learning, adaptive and other features, experiments show that the use of neural network modeling can achieve good forecast results on the stock market. Because the trend of the stock market looks chaotic, but in fact the changes are in its internal law of neural network which is the basis for forecasting the stock market. BP network makes use of the historical data of previous study to identify the inherent law of development of the stock market, and store it in the network as weighting value and the threshold value to predict the future trend, especially with more visible effect for short-term prediction.
However, BP neural network learning convergence is slow, vulnerable to the shortcomings of local minimum points, which results in the unsatisfactory prediction effect on the stock market. In view of this, this thesis adopts genetic algorithm and BP neural network integration strategy to overcome its shortcomings.
Genetic algorithm and BP neural network used in a combination of stock market forecasts are not new study, but can the turnover rate be used as a factor to forecast the stock market? How the turnover rate factor can be added into the forecasting network model? So far scholars have not done research in this aspect. As the first study, this thesis presents two kinds of turnover rate as a factor to be added to the proposal of forecasting network model: First, turnover rate and the closing price at the same time as the network input data; Second, the average turnover rate as the threshold of hidden layer value-added value. For each proposal a forecasting system is developed, running through the system to determine the actual turnover rate as a factor could be used to predict the effect of the stock market, for the purpose of researching on genetic algorithm and BP neural network convergence strategy in the application of time series prediction.
本文在深入分析股市投资理论和股价预测方法的基础上,使用BP神经网络作为股市预测的网络模型。股市市场是一个极其复杂的非线性动力学系统,而神经网络具有很强的非线性逼近能力和自学习、白适应等特性,实验证明,利用神经网络对股市建模可以取得较好的预测效果。因为股市市场的走势看起来杂乱无章,但实际上有其内在的变化规律,而这正是神经网络预测股市的基础。BP网络通过对以往历史数据的学习,找出股市运行的内在规律,并将其存储在网络具体的权值、阈值中,用以预测未来的走势,尤其对于短期的预测效果更为明显。
然而,BP神经网络存在学习收敛速度缓慢、易陷入局部极小点等缺点,使其对股市预测的效果不能令人满意。鉴于此,本文采用遗传算法与BP神经网络的融合策略,来达到克服其缺点的目的。
遗传算法与BP神经网络相结合运用于股市预测的研究早已不是什么新鲜事了,但对于换手率作为因子能否用于股市预测?如何在预测网络模型中加入换手率因子?到目前为止还没有学者作过研究。作为首次研究,本文提出了两种将换手率作为因子加入到预测网络模型的方案:一,换手率和收盘价同时作为网络的输入数据;二,平均换手率作为隐含层阈值的附加值。对每个方案开发一个预测系统,通过系统的实际运行来判断换手率作为因子能否用于股市预测以及预测效果如何,从而达到研究遗传算法与BP神经网络融合策略在时间序列预测中应用的目的。
With the economic growth and the conversion of people's investment consciousness, stock has become an important part of people's life in modern time. Investment in stock has greatly become one of focuses of public topic. How to keep the development and boost stock market is becoming the emphasis of concern and research of manager and investor. The proceeds of stock investment always equal the risk. That means the good proceeds is based on the high risk of failure. Therefore, the study of stock prediction method has great application value and theoretical significance. The complexity of inside structure and levity of exterior complication in system of stock market make stock market predication a complex problem. The traditional methods and tools have not met its challenge, the thesis presents a method of modeling stock market using neural network based on thorough study of stock investment theories and stock prediction methods.
On base of analysis of investment theory and the stock market price prediction method, BP neural network was utilized as a network model for stock market prediction. The stock market is an extremely complex non-linear dynamical system, and neural network has strong nonlinear approximation ability and self-learning, adaptive and other features, experiments show that the use of neural network modeling can achieve good forecast results on the stock market. Because the trend of the stock market looks chaotic, but in fact the changes are in its internal law of neural network which is the basis for forecasting the stock market. BP network makes use of the historical data of previous study to identify the inherent law of development of the stock market, and store it in the network as weighting value and the threshold value to predict the future trend, especially with more visible effect for short-term prediction.
However, BP neural network learning convergence is slow, vulnerable to the shortcomings of local minimum points, which results in the unsatisfactory prediction effect on the stock market. In view of this, this thesis adopts genetic algorithm and BP neural network integration strategy to overcome its shortcomings.
Genetic algorithm and BP neural network used in a combination of stock market forecasts are not new study, but can the turnover rate be used as a factor to forecast the stock market? How the turnover rate factor can be added into the forecasting network model? So far scholars have not done research in this aspect. As the first study, this thesis presents two kinds of turnover rate as a factor to be added to the proposal of forecasting network model: First, turnover rate and the closing price at the same time as the network input data; Second, the average turnover rate as the threshold of hidden layer value-added value. For each proposal a forecasting system is developed, running through the system to determine the actual turnover rate as a factor could be used to predict the effect of the stock market, for the purpose of researching on genetic algorithm and BP neural network convergence strategy in the application of time series prediction.
引文
[1]刘莉华.神经网络方法在股市预测中的应用研究[D].成都:电子科技大学,2005
[2]马向前,万帼荣.影响我国股票市场价格波动的基本因素[J].山西统计,2001(1):22-23
[3]陈之大,贺学会.证券投资技术分析[M].西南财经大学出版社.1996
[4]吴成东,王长涛.人工神经元BP网络在股市预测方面的应用[J].控制工程,2002,27(3):34-37
[5]张玉林,吴微.用BP神经网络捕捉股市黑马初探.运筹与管理,2004,35(2):23-26
[6]王小平,曹立明.遗传算法—理论、应用与软件实现[M].西安:西安交通大学出版社,2002
[7]张文修,梁怡.遗传算法的数学基础[M].西安:西安交通大学,2000:14-35
[8]Bennett,K.A genetic algorithm for database query optimization[J].IEEE conference on Evolutionary Computation,1994:400-407
[9]Bagley J D.The Behavior of Adaptive System which Employ Genetic and Correlation Algorithm[J].Dissertation Abstracts Intenational,1967
[10]孟祥武,程虎.进化神经网络编码表示机制研究[J].计算机科学,1998,25(1):55-57
[11]黄景平,蔡骏,冯珊.基于agent的神经网络训练模式[J].华中理工大学学报,1999,27(3):78-79
[12]程烨.遗传算法在路径覆盖测试数据生成中的研究与应用[D].上海师范大学,2006
[13]韩万林,张幼蒂.遗传算法的改进[J].中国矿业大学学报,2000(1):12-14
[14]任咏红.具有模糊关系不等式约束的优化问题[D].辽宁师范大学,2003
[15]黄景平,蔡骏,冯珊.基于agent的神经网络训练模式[J].华中理工大学学报.1999,27(3):78-79
[16]王梦兰.混合式遗传算法及其在指派问题与排序问题中的应用[D].西南交通大学,2002
[17]程启明,万德均,陈雪丽.遗传学习算法的神经网络自适应船舶操纵控制系统研究[J].模式识别与人工智能.1998,11(3):305-309
[18]郑丽君.基于遗传算法的多目标优化与决策方法研究[D].国防科学技术大学,2003
[19]方咸云.基于遗传算法的智能白适应控制系统研究[D].华侨大学,2003
[20]周翔,蔡自兴.基于多分辨率遗传算法的多层感知神经网络及其在短期电力负荷预估中的应用[J].计算机工程与应用,1997(10):10-12
[21]李军.基于最优基因的遗传算法研究[D].曲阜师范大学,2007
[22]Montana,D.J.and Davis,L.Training feed-forward neutral networks using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial Intelligence,Los Altos,CA:Morgan Kaufmann Publishers,1989:762-767
[23]Whitely D.Fundamental Principles of Deception in Genetic Search[A].In:Foundations of Genetic Algorithms.Morgan Kaufmann Publishers,1989:221-241.
[24]高隽编著.人工神经网络原理及仿真实例[M].北京:机械工业出版社,2003
[25]王旭,王宏等编著.人工神经网络原理与应用[M].辽宁:东北大学出版社,2000
[26]吴仁志.单神经元和多层前向人工神经网络的研究与应用[D].广西大学,2002
[27]Ormoneit D,Neuneier R.Experiments in Predicting the Germ an Stock Index DAX with Density Estimating Neural networks.QIEEE/IAFE 1996 Conference on Computational Intelligence for Financial Engineering.1996:66-77
[28]Montana,D.J.and Davis,L.Training feed-forward neutral networks using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial intelligence,Los Altos,CA:Morgan Kaufmann Publishers.1989:762-767
[29]Refenes A N,Zapranis A,Francis G.Stock performance modeling using neural networks:a comparative study with regression models[J].Neural Networks.1994,7(2):375-388
[30]Montana,D.J.and Davis,L.Training feed-forward neutral networks,using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial Intelligence,Los Altos,CA:Morgan Kaufmann Publishers.1989:762-767.
[31]Xin Yao.A review of evolutionary artifical neural network[J].International Journal of Intelligent Systems.1993,8(3):539-565
[32]郭光,严绍谨.人工神经网络用于我国东部汛期降水预测研究[J].南京气象学院学报,1996,19(3):354-357
[33]张青贵.人工神经网络导论[M].北京:中国水利水电出版社,2004
[34]李晓峰.BP人工神经网络模型的进一步改进及其应用[J].吉林化工学院学报,2000,17(4):48-51
[35]张玉林,吴微.用BP神经网络捕捉股市黑马初探[J].运筹与管理,2004,13(2):123-126
[36]刘永福,李建功.利用BP神经网络预测上证指数[J].金融观察,2003(10):87-91
[37]陈可,张琴舜.BP神经网络在证券分析预测中应用[J].计算机工程,2001(11):95-97
[38]张立明.人工神经网络的模型及其应用[D].上海:复旦大学,1993
[39]禹建丽,孙增圻.基于BP神经网络的股市建模与决策[J].系统工程理论与实践.2003,37(5):17-21
[40]Holland J.H.Adaptation in Natural and Artificial Systems[M].Ann Arbor:University of Michigan press.1975
[41]张军英,保铮.前向网络隐层节点数的最小上界研究[C].中国神经计算科学大会论文集.北京:人民邮电出版社,1997:177-184
[42]刘建娟,徐晓苏.改进遗传人工神经网络在组合导航中的应用[J].中国惯性技术学报,2006,14(5):24-27
[43]倪远平,邹金慧,陈艾.改进型GA-BP训练策略与模式分类[J].小型微型计算机系统.1997,19(7):40-44
[44]梁化楼,戴贵亮.人工神经网络与遗传算法的结合:进展及展望[J].电子学报,1995,(10):195-200
[45]懂聪,郭晓华.计算智能中若干热点问题的研究与进展[J].控制理论与应用,2000,17(5):691-697
[46]董聪.多层前向网络的逼近与泛化机制[J].控制与决策.1998,13(4):414-417
[47]佚名.遗传算法研究方向[EB/OL].中国人工智能网.2007-02.http://www.intsci.ac.cn/intro/direction.html
[48]刘永红.神经网络理论的发展与前沿问题[EB/OL].科研中国2006-04-04.http://www.sciei.com/Article/Com/computer/200604/Article-5421-6.html
[49]赵云珍.遗传算法及其改进[D].昆明理工大学,2005
[50]王熙.基于遗传算法改进的BP神经网络在矿产品预测领域的应用[D].北京:中国地质大学,2006
[2]马向前,万帼荣.影响我国股票市场价格波动的基本因素[J].山西统计,2001(1):22-23
[3]陈之大,贺学会.证券投资技术分析[M].西南财经大学出版社.1996
[4]吴成东,王长涛.人工神经元BP网络在股市预测方面的应用[J].控制工程,2002,27(3):34-37
[5]张玉林,吴微.用BP神经网络捕捉股市黑马初探.运筹与管理,2004,35(2):23-26
[6]王小平,曹立明.遗传算法—理论、应用与软件实现[M].西安:西安交通大学出版社,2002
[7]张文修,梁怡.遗传算法的数学基础[M].西安:西安交通大学,2000:14-35
[8]Bennett,K.A genetic algorithm for database query optimization[J].IEEE conference on Evolutionary Computation,1994:400-407
[9]Bagley J D.The Behavior of Adaptive System which Employ Genetic and Correlation Algorithm[J].Dissertation Abstracts Intenational,1967
[10]孟祥武,程虎.进化神经网络编码表示机制研究[J].计算机科学,1998,25(1):55-57
[11]黄景平,蔡骏,冯珊.基于agent的神经网络训练模式[J].华中理工大学学报,1999,27(3):78-79
[12]程烨.遗传算法在路径覆盖测试数据生成中的研究与应用[D].上海师范大学,2006
[13]韩万林,张幼蒂.遗传算法的改进[J].中国矿业大学学报,2000(1):12-14
[14]任咏红.具有模糊关系不等式约束的优化问题[D].辽宁师范大学,2003
[15]黄景平,蔡骏,冯珊.基于agent的神经网络训练模式[J].华中理工大学学报.1999,27(3):78-79
[16]王梦兰.混合式遗传算法及其在指派问题与排序问题中的应用[D].西南交通大学,2002
[17]程启明,万德均,陈雪丽.遗传学习算法的神经网络自适应船舶操纵控制系统研究[J].模式识别与人工智能.1998,11(3):305-309
[18]郑丽君.基于遗传算法的多目标优化与决策方法研究[D].国防科学技术大学,2003
[19]方咸云.基于遗传算法的智能白适应控制系统研究[D].华侨大学,2003
[20]周翔,蔡自兴.基于多分辨率遗传算法的多层感知神经网络及其在短期电力负荷预估中的应用[J].计算机工程与应用,1997(10):10-12
[21]李军.基于最优基因的遗传算法研究[D].曲阜师范大学,2007
[22]Montana,D.J.and Davis,L.Training feed-forward neutral networks using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial Intelligence,Los Altos,CA:Morgan Kaufmann Publishers,1989:762-767
[23]Whitely D.Fundamental Principles of Deception in Genetic Search[A].In:Foundations of Genetic Algorithms.Morgan Kaufmann Publishers,1989:221-241.
[24]高隽编著.人工神经网络原理及仿真实例[M].北京:机械工业出版社,2003
[25]王旭,王宏等编著.人工神经网络原理与应用[M].辽宁:东北大学出版社,2000
[26]吴仁志.单神经元和多层前向人工神经网络的研究与应用[D].广西大学,2002
[27]Ormoneit D,Neuneier R.Experiments in Predicting the Germ an Stock Index DAX with Density Estimating Neural networks.QIEEE/IAFE 1996 Conference on Computational Intelligence for Financial Engineering.1996:66-77
[28]Montana,D.J.and Davis,L.Training feed-forward neutral networks using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial intelligence,Los Altos,CA:Morgan Kaufmann Publishers.1989:762-767
[29]Refenes A N,Zapranis A,Francis G.Stock performance modeling using neural networks:a comparative study with regression models[J].Neural Networks.1994,7(2):375-388
[30]Montana,D.J.and Davis,L.Training feed-forward neutral networks,using genetic algorithms [J].In Proceeding of the International Joint Conference on Artificial Intelligence,Los Altos,CA:Morgan Kaufmann Publishers.1989:762-767.
[31]Xin Yao.A review of evolutionary artifical neural network[J].International Journal of Intelligent Systems.1993,8(3):539-565
[32]郭光,严绍谨.人工神经网络用于我国东部汛期降水预测研究[J].南京气象学院学报,1996,19(3):354-357
[33]张青贵.人工神经网络导论[M].北京:中国水利水电出版社,2004
[34]李晓峰.BP人工神经网络模型的进一步改进及其应用[J].吉林化工学院学报,2000,17(4):48-51
[35]张玉林,吴微.用BP神经网络捕捉股市黑马初探[J].运筹与管理,2004,13(2):123-126
[36]刘永福,李建功.利用BP神经网络预测上证指数[J].金融观察,2003(10):87-91
[37]陈可,张琴舜.BP神经网络在证券分析预测中应用[J].计算机工程,2001(11):95-97
[38]张立明.人工神经网络的模型及其应用[D].上海:复旦大学,1993
[39]禹建丽,孙增圻.基于BP神经网络的股市建模与决策[J].系统工程理论与实践.2003,37(5):17-21
[40]Holland J.H.Adaptation in Natural and Artificial Systems[M].Ann Arbor:University of Michigan press.1975
[41]张军英,保铮.前向网络隐层节点数的最小上界研究[C].中国神经计算科学大会论文集.北京:人民邮电出版社,1997:177-184
[42]刘建娟,徐晓苏.改进遗传人工神经网络在组合导航中的应用[J].中国惯性技术学报,2006,14(5):24-27
[43]倪远平,邹金慧,陈艾.改进型GA-BP训练策略与模式分类[J].小型微型计算机系统.1997,19(7):40-44
[44]梁化楼,戴贵亮.人工神经网络与遗传算法的结合:进展及展望[J].电子学报,1995,(10):195-200
[45]懂聪,郭晓华.计算智能中若干热点问题的研究与进展[J].控制理论与应用,2000,17(5):691-697
[46]董聪.多层前向网络的逼近与泛化机制[J].控制与决策.1998,13(4):414-417
[47]佚名.遗传算法研究方向[EB/OL].中国人工智能网.2007-02.http://www.intsci.ac.cn/intro/direction.html
[48]刘永红.神经网络理论的发展与前沿问题[EB/OL].科研中国2006-04-04.http://www.sciei.com/Article/Com/computer/200604/Article-5421-6.html
[49]赵云珍.遗传算法及其改进[D].昆明理工大学,2005
[50]王熙.基于遗传算法改进的BP神经网络在矿产品预测领域的应用[D].北京:中国地质大学,2006