基于RS-LS-SVM的股票市场预测模型研究

发布时间：2018-12-29 16:53

【摘要】：支持向量机作为一种常见的数据挖掘方法,相对其他方法来说具有十分突出的优点,目前在各个领域的应用也十分广泛。但是,该方法在实际运用中还有许多问题值得研究,模型本身也有很多可以改进的地方。为了进一步地改进支持向量机模型并对其进行推广,本文选取了目前研究较多,但对支持向量机方法的运用上还存在问题的资本市场作为研究对象。本文针对目前的股市预测方法以及标准支持向量机存在的缺陷进行了分析,在此基础上提出了结合粗糙集的最小二乘支持向量机模型对股市进行预测,首先应用粗糙集对预测指标进行属性约简,然后对约简的指标体系使用最小二乘支持向量机来预测股价的波动情况,以期能够给股市预测提供一定的参考,并对粗糙集与最小二乘支持向量机的应用提供思路与方法。本文的主要研究内容为:首先系统地介绍了关于股票市场预测、支持向量机、最小二乘支持向量机和粗糙集的国内外研究现状,并在前人研究的基础上总结了现有方法的不足,鉴于现有方法的不足,提出RS-LS-LSVM股市预测方法;其次对粗糙集和最小二乘支持向量机的基础理论以及模型中涉及的核函数的选择问题进行了说明;随后根据模型的整体思路建立了模型的流程图,详细描述了该预测模型的处理过程,并建立了一套包括今日最高价、昨日最高价、前日最高价、7日平均最高价等在内的27个指标构成的预测指标体系,并对模型中各个步骤的处理过程与方法进行了一一说明;最后在主板、中小板、创业板中随机选取了中国石油(601857)、辉隆股份(002556)和劲胜精密(300083)的2016年全年的交易数据作为研究样本,每个样本都是244组数据,在MATLAB软件中进行了三次三组模型的对比实验,每次对比实验又进行了20次的随机试验,在软件中分别使用RS-LS-SVM、LS-SVM和RS-SVM对样本数据回归预测,并对结果进行了对比。实验结果表明:一、实验针对三个样本得到了三个不同的约简指标体系,说明相同的指标在针对不同的预测对象时,有效指标不同;相同的指标对不同的预测对象的应用上也存在区别。所以预测之前进行指标筛选和属性约简是十分必要的,属性约简可以减少数据冗余,提高预测性能,并且本文提出的初选指标体系能够在一定程度上对我国股市进行预测。二、实验结果验证了RS-LS-SVM预测模型的可行性和有效性。无论是在主板、中小板还是在创业板中,多次实验结果表明,在MSE和RMSE两个数据上表现,RS-LS-SVM预测模型都比LS-SVM和RS-SVM模型更为优秀。可以看出,将粗糙集与最小二乘支持向量机引入股票市场中,简化了模型难度,提高了求解速度,具有一定的创新性,这对于股票市场预测模型的研究和投资者进行投资决策具有参考价值。
[Abstract]:As a common data mining method, support vector machine (SVM) has many advantages compared with other methods, and it is widely used in various fields. However, there are still many problems worth studying in practical application, and the model itself can be improved. In order to further improve the support vector machine (SVM) model and generalize it, this paper selects the capital market which has some problems in the application of support vector machine (SVM). Based on the analysis of the current stock market forecasting methods and the defects of the standard support vector machine (SVM), a rough set least squares support vector machine (LS-SVM) model is proposed to predict the stock market. First, the rough set is used to reduce the attribute of the prediction index, and then the least square support vector machine is used to predict the fluctuation of stock price in the index system of the reduction, so as to provide a certain reference for the stock market forecast. It also provides ideas and methods for the application of rough set and least squares support vector machine. The main contents of this paper are as follows: firstly, this paper systematically introduces the research status of stock market prediction, support vector machine, least square support vector machine and rough set, and summarizes the shortcomings of existing methods on the basis of previous studies. In view of the shortcomings of the existing methods, the RS-LS-LSVM stock market forecasting method is put forward. Secondly, the basic theory of rough set and least squares support vector machine (LS-SVM) and the selection of kernel function in the model are explained. Then, according to the overall idea of the model, the flow chart of the model is established, the processing process of the prediction model is described in detail, and a set of prices including the highest price today, the highest price yesterday, the highest price the day before yesterday, The prediction index system is composed of 27 indexes including the highest price of 7 days, and the processing process and method of each step in the model are explained one by one. Finally, in the main board, the small and medium-sized board, and the gem, the transaction data of PetroChina (601857), Huilong shares (002556) and Jinsheng Precision (300083) for the whole year 2016 were randomly selected as the research samples. Each sample was 244sets of data. The contrast experiments of three groups of models were carried out in MATLAB software, and 20 random experiments were carried out each time. In the software, RS-LS-SVM,LS-SVM and RS-SVM were used to predict the sample data respectively. The results are compared. The experimental results show that: first, three different reduction index systems are obtained for the three samples, which shows that the same index is different for different prediction objects; There are differences in the application of the same indicators to different prediction objects. Therefore, it is very necessary to carry out index selection and attribute reduction before prediction. Attribute reduction can reduce data redundancy and improve prediction performance, and the primary index system proposed in this paper can predict the stock market in China to a certain extent. Second, the experimental results verify the feasibility and validity of the RS-LS-SVM prediction model. The results of many experiments on the main board, the small and medium-sized board and the growth enterprise board show that the RS-LS-SVM prediction model is better than the LS-SVM and RS-SVM model in the data of MSE and RMSE. It can be seen that introducing rough set and least squares support vector machine into stock market simplifies the difficulty of the model and improves the speed of solution. It has reference value for the research of stock market forecasting model and investors' investment decision.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】