联系我们

论文查询结果	返回搜索

论文编号：	3264
作者编号：	2120092275
上传时间：	2011/6/6 16:35:35
中文题目：	基于主成分分析与神经网络的股票价格预测
英文题目：	Predict the stock price using principle component analysis and neural network
指导老师：	安利平
中文关键字：	BP神经网络,主成分分析,股票价格,预测
英文关键字：	BP neural network, principle component analysis, stock price, prediction
中文摘要：	股市是市场经济的必然产物，在一个国家的金融领域有着举足轻重的地位。随着证券市场的逐步规范，人们的投资理念不断提高，股票投资己成为众多家庭及个人理财的一种重要方式。然而，股票市场具有高风险与高收益并存的特性，研究股票价格的运行规律对于理性投资具有重要的作用。 进行股票价格预测，首先必须承认股票市场的运行存在相对稳定的规律。这些规律全都隐藏在股票价格运行的历史数据中。从数学的角度讲，这种规律表现为一种复杂的函数关系。预测的目的就是要找出这种规律，从而利用这种规律获得对股票价格未来的发展的预先把握。由于神经网络具有很强的函数逼近能力，能够在训练的过程中学习到隐含在样本中的信息，并且将其记录在网络的权值和阈值中。因此人工神经网络用于股价预测将会有很大优势。 股价预测要求神经网络有较好的泛化能力。网络泛化性能的好坏取决于人工神经网络是否从训练样本中找到隐含的真正规律。从BP神经网络运行机制的分析可以看出，神经网络的泛化性能主要受两个方面因素的影响。第一，训练样本的影响，包括样本的质量、数量和代表性。第二，网络结构的复杂性。 本文在分析考察传统预测分析方法的基础上，针对影响网络泛化性能的两个因素提出一个由BP神经网络和主成分分析方法相结合的预测方法，并对该方法泛化性能的改善和提高进行了深入研究。将主成分分析法引入股价预测，应用SPSS对原始输入变量进行预处理，并用选择的主成分代替原始输入变量作为神经网络模型新的输入变量。经过这样的处理后，一方面能够减少输入变量的维度，并消除了各输入变量之间的线性相关性；另一方面由于神经网络的结构与输入层的变量个数密切相关，减少输入变量使神经网络的结构得到简化，提高了神经网络的收敛性和稳定性，并且提高的网络的泛化能力。实验选取了中国石化的股票数据进行建模研究，并且将传统BP网络与基于主成分分析的BP网络的预测效果进行了对比，发现后者的预测精度更高。之后，本文选择了上证综合指数进行了对比实验，验证了该方法的可扩展性。
英文摘要：	The stock market is the inevitable product of market economy, and it has a pivotal position in a country's economy. As the stock market is gradually standardized, people's investment philosophy is improving, and stock investment has become an important way of money management for personal finance or organizational finance. However, the stock market is characterized by high risk and high yield, so the stock price’s changes and trends have widely received the attention of the government and the public. In order to forecast the stock price, we must recognize that the running of stock market has its own mechanism which is hidden in the historical data. From the point of mathematics, the mechanism can be expressed as a complex function. The purpose of forecasting the stock price is to discovery this mechanism and find out the trend of stock price. With the strong ability of function approximation, the neural network can discovery the information hidden in the samples and record it in the network’s weights and threshold values during the training process. Therefore, there will be a great advantage to predict stock price with the artificial neural networks. Neural network for stock price prediction requires excellent generalization ability. The network’s generalization performance depends on whether the artificial neural network can find the true law that is hidden in the training samples. It has been discovered from BP neutral network’s mechanism that the generalization performance of neural networks is mainly affected by two factors. The first is the training samples, including the quality, quantity and representation of them. The second is the complexity of network’s structure. Based on the analysis of traditional forecasting methods, a combined forecasting method is proposed which based on a neural network and principal component analysis. Furthermore, a deep research has been done to improve the generalization performance of the network. This paper introduces the principal component analysis into the stock price forecasting, preprocess the original input variables of the neural network, and select the principal components of the input variables as the new input of network model. On the one hand, it will reduce the dimension of the input, and eliminate the linear correlation of the input variables. On the other hand, it simplifies the structure of the neural network and then improve the convergence and stability of the neural network and improve the network’s generalization. Sinopec’s stock data were chosen for modeling, and compared the outcome of the BP network and the PCA-based BP neural network. The result showed that the prediction accuracy of the latter was higher. Afterwards, this paper chose the Shanghai composite index to make a comparative experiment in order to test the scalability of the method.
查看全文：	预览  下载(下载需要进行登录)