基于信息粒化的SVM在证券时间序列分析中的应用

发布时间：2018-03-23 19:08

本文选题：时间序列　切入点：非线性　出处：《昆明理工大学》2014年硕士论文

【摘要】：随着经济的快速发展,金融领域的研究越来越受到国内外学者的重视。近年来时间序列分析的方法有好多种,传统的模型要么模型非线性映射比较简单,不能完全反映非线性规律；要么模型结构复杂,推广性不强。支持向量机(Support Vector Machines,SVM)提供了较好的建模方法,有较好的理论基础,不需要考虑系统的数学模型,对数据拟合采用结构风险最小化,训练率高。本文将信息粒化与SVM组合,对证券时间序列进行回归分析。首先对时间序列分析、SVM和信息粒化的理论进行了比较深入的介绍,并将SVM运用到时间序列的分析当中,将证券时间序列的数据根据模型进行自变量和因变量的选择,并对这些数据进行归一化处理；其次针对SVM相关的参数进行调节的问题进行了研究,主要介绍了三种参数寻优方法,分别是交叉验证法、遗传算法、微粒群优化算法,对这些算法的效率进行比较：最后对上证指数进行模糊信息粒化后再进行SVM的回归分析,这样对上证收盘指数的变化趋势和空间进行了有效的预测。实验结果表明SVM模型很好的反映了上证指数的变化规律,拟合和预测还是比较理想的：同时将信息粒化和SVM模型的有效结合可以使SVM发挥更好的效果,而且这种方法是十分可行的。
[Abstract]:With the rapid development of economy, the financial sector research more and more attention by scholars. There are a lot of methods of time series analysis in recent years, the traditional model or model of nonlinear mapping is relatively simple, can not fully reflect the nonlinear law; or the complex structure of the model, the promotion is not strong. The support vector machine (Support Vector Machines. SVM) provides a good modeling method, a good theoretical basis, do not need to consider the mathematical model of the system, the data fitting using structural risk minimization, the training rate is high. The information granulation in combination with SVM regression analysis for stock time series.
First, analysis of the time series, SVM and information granulation theory this paper introduces, and the SVM is applied to the analysis of time series, the stock time series data according to the model of the independent and dependent variables, and these data are normalized processing parameters; secondly, SVM related regulation the problem is studied, this paper introduces three kinds of optimization methods, namely the method of cross validation, genetic algorithm, particle swarm optimization algorithm, compare the efficiency of the algorithm. In the end, the stock index regression analysis of fuzzy information granulation after SVM, which was effective in predicting the change trend and the Shanghai index closed space.
The experimental results show that the SVM model reflects the change rule of the Shanghai stock index very well, and the fitting and prediction is ideal. At the same time, the effective combination of information granulation and SVM model can make SVM play a better effect, and this method is very feasible.

【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F830.91;TP18

【参考文献】