基于改进动态神经网络的股票预测模型的研究

发布时间：2018-03-29 05:30

本文选题：动态神经网络　切入点：股票预测　出处：《内蒙古大学》2014年硕士论文

【摘要】：人工神经网络理论应用于股票走势预测,就是利用由股票的历史交易数据组成的时间序列,通过神经网络的自学习能力对其进行分析,从复杂的数据关系中找出其中的规律,然后模拟网络输出数据与输入数据之间的函数关系,并将此函数用于对未来股票价格的预测中。近年来,国内外研究学者已经基于神经网络理论建立了多种预测模型对股票价格走势进行预测,并取得了显著的研究成果。但综观目前的股票预测模型,大多是基于静态神经网络理论,如BP神经网络、Wavelet神经网络和RBF神经网络等。然而,这些静态神经网络预测模型并不能充分反映股市系统的动态特性。本文通过分析现有的股票预测模型在动态性方面的不足,以及输入指标选取方面存在的问题,提出了基于主成分分析法的动态神经网络股票预测模型。该模型通过建立多步输入输出时延和输出反馈机制,更好的描述了股票市场这个动态时变系统的特性。并为预测指标的选取和指标个数的确定提供了一个新的方法。模型建立后,本研究选取了多只股票的历史数据作为实验对象,并对其进行了多次的实验,然后将得出的实验结果与传统的BP神经网络模型进行了比较。实验结果表明,本文提出的基于主成分分析法的动态神经网络模型是有效的。
[Abstract]:The application of artificial neural network theory to stock trend prediction is to use the time series composed of historical trading data of stock, to analyze it through the self-learning ability of neural network, and to find out the rules from the complex data relationship. Then the functional relationship between the network output data and the input data is simulated, and the function is used to predict the future stock price. In recent years, researchers at home and abroad have established a variety of forecasting models based on the neural network theory to forecast the trend of stock prices, and achieved remarkable research results. Most of them are based on static neural network theory such as BP neural network wavelet neural network and RBF neural network. However these static neural network prediction models can not fully reflect the dynamic characteristics of stock market system. This paper analyzes the shortage of the existing stock forecasting model in the dynamic aspect, and the problems in the selection of the input index. A dynamic neural network stock prediction model based on principal component analysis (PCA) is proposed, which establishes a multi-step input-output delay and output feedback mechanism. A better description of the characteristics of the dynamic time-varying system of the stock market is given, and a new method is provided for the selection of prediction indicators and the determination of the number of indicators. After the establishment of the model, the historical data of several stocks are selected as the experimental objects. The experimental results are compared with the traditional BP neural network model. The experimental results show that the proposed dynamic neural network model based on principal component analysis is effective.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;TP183

【参考文献】