基于图像处理方法的股票数据分析研究

发布时间：2018-03-23 03:36

本文选题：股票板块　切入点：相关性　出处：《重庆大学》2012年硕士论文　论文类型：学位论文

【摘要】：目前，人们的金融意识的日益增强，引起越来越多的投资者对股票投资的青睐。因此，对股票市场的分析和预测的研究就越有其必要性，研究者们也一直致力于利用各种方法对某支股票、某股票指数或不同板块股票的走势的预测研究。虽然股票的波动可能为投资者们带来不确定的收益，可由于其影响的因素较多，使得投资者们难以并且也不可能完全掌握股票确切的涨跌规律。但若能估计股票的基本涨跌情况，一定程度上也可以给投资者一些建议。对板块内的所有股票数据间的研究，有利于在了解板块整体的涨跌情况以及板块内是否存有涨势情况异于其他的股票后，为投资者对该板块的后期发展的分析提供一些参考信息。因此，有关估计板块股票涨跌趋势的研究具有一定的价值。投资者一直都希望能够掌握变化莫测的股市的涨跌规律，从而出现了很多技术分析方法以及股票预测方法，其中技术分析方法包含主观成分较多，不同的研究者的结论可能会有一定的出入。股票预测方法主要有数学模型和无模型两类。传统的关于股票数据分析预测的研究，主要是侧重对单支股票某些股票的研究，通过指定的模型对数据做实证分析。在建立数学模型时，主要选取适当的变量，以保证做实证分析预测时有一定的可靠性和准确性，但是建立数学模型时要检验各个变量的显著性以及考虑变量的合适性，从而计算量比较大，并不具有直观性。本文的研究内容是结合图像处理方法对股票数据进行分析。首先将收集到的某板块股票数据做归一化处理，，再形成灰度图，图像的竖直方向表示板块中的不同股票，而水平方向表示处理后的股票不同日期的收盘价格数据。由于各支股票间的相关性强弱将影响图像在竖直方向上的连贯性和光滑性，因此需对该板块内的股票重新排列。投资者主要关注股票的大致走势，忽略股票间那些小的波动，这些小的波动可视为噪声。在分析由股票板块数据形成的灰度图时，噪声将影响图像的清晰度。文章中结合股票数据图像的特点，将均值滤波法、中值滤波法和自适应维纳（Wiener）滤波法分别对图像进行竖直方向上去噪。将实验结果进行对比，同时结合处理后的股票数据图像对该板块的涨势进行了分析，进而能得出该板块是否具有明显的板块效应。最后，对去噪后的图像分别进行水平方向和竖直方向的边缘提取。提取的竖直方向边缘主要是反映了该股票板块在某时刻的涨跌幅度的极大值或极小值；而水平方向的边缘主要反映了该板块内某支股票涨势的奇异性。提取边缘所用的方法主要是利用小波变换，找到数据中的极值，从而找到对应方向上的边缘。结合这些边缘所处的位置间隔分析该板块的大致涨跌周期。与既往的方法相比，本文从二维的角度来分析股票数据且能够较直观的反映该板块股票的整体涨跌趋势。
[Abstract]:At present, with the increasing of people's financial consciousness, more and more investors prefer to invest in stocks. Therefore, it is necessary to study the analysis and prediction of stock market. Researchers have also been using a variety of methods to predict the movements of a particular stock, a stock index, or a different sector, although volatility in stocks can bring uncertain returns to investors. However, because of its many factors, it makes it difficult and impossible for investors to fully grasp the exact rise and fall rules of the stock. But if we can estimate the basic rise and fall of the stock, To a certain extent, some suggestions can also be given to investors. The study of all the stock data in the plate is helpful in understanding the overall rise and fall of the plate and whether there is a rise in the plate that is different from that of other stocks. This paper provides some reference information for investors to analyze the late development of the plate. Therefore, the research on estimating the trend of stock price rise and fall of the plate is of certain value. Investors have always hoped to be able to master the fluctuating rules of the stock market. As a result, there have been many technical analysis methods and stock forecasting methods, among which the technical analysis methods contain more subjective elements. The conclusions of different researchers may be different. There are two main methods of stock prediction: mathematical model and no model. The traditional research on stock data analysis and prediction mainly focuses on the research of some stocks in a single stock. When establishing the mathematical model, the appropriate variables are selected to ensure the reliability and accuracy of the empirical analysis and prediction. But it is necessary to test the significance of each variable and the appropriateness of considering the variable when establishing the mathematical model so that the calculation is large and not intuitive. The research content of this paper is to analyze the stock data with image processing method. Firstly, the stock data collected from a certain plate is normalized, and then a gray map is formed, and the vertical direction of the image represents different stocks in the block. The horizontal direction represents the closing price data of the processed stocks on different dates. Because the correlation between the stocks will affect the coherence and smoothness of the image in the vertical direction, Investors focus on the general trend of stocks, ignoring the small fluctuations between stocks, which can be regarded as noise. When analyzing grayscale maps formed by stock plate data, The noise will affect the sharpness of the image. In this paper, mean filter, median filter and adaptive Wiener filter are used to de-noise the image in vertical direction according to the characteristics of stock data image, and the experimental results are compared. At the same time, combined with the processed stock data image, we analyze the rise of the plate, and then we can find out whether the plate has obvious plate effect. Finally, The edge of the image after denoising is extracted in horizontal direction and vertical direction respectively. The extracted vertical edge mainly reflects the maximum or minimum value of the stock plate's rise and fall at a certain time. The edge of the horizontal direction mainly reflects the singularity of a stock rise in the plate. The method to extract the edge is to use wavelet transform to find the extreme value in the data. In order to find the corresponding edge in the direction. Combined with the position interval of these edges to analyze the roughly rising and falling cycles of the plate. Compared with the previous methods, This paper analyzes the stock data from a two-dimensional perspective and can directly reflect the overall trend of the stock market.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F830.91

【参考文献】