基于时间序列相似性的股价趋势预测研究

发布时间：2018-01-18 08:38

本文关键词：基于时间序列相似性的股价趋势预测研究　出处：《重庆交通大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着计算机技术应用范围逐步扩大，时间序列数据频繁出现于交通、商业、科学、金融等各个领域，时间序列数据分析处理技术被越来越多人重视。传统时间序列分析预测方法往往将时间序列数据匹配到某些数学模型中，然后再对其整体进行分析和预测，但是现实中许多数据不能够满足模型参数要求。针对于此，基于时间序列相似性的类比合成预测方法以其非参数回归特性成为该领域研究焦点。本文在对时间序列相似性度量方法和类比合成方法的研究基础上，提出了一种不等长时间序列相似性度量方法，并设计出了适用性较强的时间序列趋势预测方案，以真实股票价格数据为基础进行了实证分析，主要工作及创新点如下：第一，对时间序列相似性度量方法做深入研究，并对时间序列中经常出现的振幅平移、振幅伸缩、线性漂移和时间轴伸缩等形变做详细讨论，认为优秀的时间序列相似性度量方法应该对上述形变不敏感。第二，对时间序列预测的类比合成方法进行深入研究，并对非参数回归模型做相关讨论，类比合成方法作为一种典型的非参数回归方法，，具有良好的应用前景。第三，提出改进型余弦公式的不等长时间序列相似性度量（RCBS_UL）算法，该算法在原始余弦公式的基础上通过对序列进行等长化处理、归一化处理，最终实现不等长时间序列的相似性度量，该方法对振幅平移、振幅伸缩、线性漂移和时间轴伸缩不敏感。第四，将时间序列预测的类比合成方法和RCBS_UL算法相结合，设计出一种时间序列趋势预测方案，并以真实股票价格指数为基础实验数据，对股票价格走势进行预测。实验结果表明，该方案能够准确预测股价走势方向，但对于未来值的确切预测还不能令人满意，需要做进一步的研究和改进。第五，结合RCBS_UL算法和相似搜索技术，设计出一种决策支持方案，该方案并没有对股票价格走势做预测计算，而是通过搜索并提供经典图形给使用者来支持其预测判断，从而帮助使用者减轻负担。
[Abstract]:With the gradual expansion of the application of computer technology, time series data frequently appear in traffic, commerce, science, finance and other fields. Time series data analysis and processing technology has been paid more and more attention. Traditional time series analysis and prediction methods often match time series data to some mathematical models and then analyze and predict the whole time series data. However, many data can not meet the requirements of model parameters in reality. In view of this, analogical composite prediction method based on similarity of time series has become the focus of research in this field because of its non-parametric regression characteristics. On the basis of the research of similarity measurement and analogical composition of time series, an unequal time series similarity measurement method is proposed in this paper. And designed a more applicable time series trend prediction scheme, based on the real stock price data for empirical analysis, the main work and innovation as follows: First, the similarity measurement method of time series is deeply studied, and the deformation such as amplitude translation, amplitude stretching, linear drift and time axis stretching are discussed in detail. It is considered that the excellent time series similarity measurement method should be insensitive to the above deformation. Secondly, the analogue synthesis method of time series prediction is studied in depth, and the non-parametric regression model is discussed. As a typical non-parametric regression method, analogue synthesis method is used as a typical non-parametric regression method. It has good application prospect. Thirdly, the RCBSULL algorithm of the improved cosine formula is proposed, which is based on the original cosine formula and is processed by equal-length processing on the basis of the original cosine formula. The method is not sensitive to amplitude translation, amplitude stretching, linear drift and time axis scaling. In 4th, combining the analog synthesis method of time series prediction with RCBS_UL algorithm, we designed a time series trend prediction scheme, and based on the real stock price index as the experimental data. The experimental results show that the scheme can accurately predict the direction of stock price trend, but the exact prediction of the future value is not satisfactory, which needs further study and improvement. 5th, combined with RCBS_UL algorithm and similar search technology, a decision support scheme is designed, which does not predict the stock price trend. It helps users to lighten their burden by searching and providing classical graphics to users to support their prediction judgment.
【学位授予单位】：重庆交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F830.91;F224

【参考文献】