基于PCA-MLR-MEABP模型的上证指数预测

发布时间：2018-05-19 21:01

本文选题：技术分析指标 + 主成分分析(PCA)　；参考：《兰州大学》2017年硕士论文

【摘要】：在股票市场的发展过程中,人们一直致力于如何准确地预测股票价格,与此同时,预测方法也在不断更新。到目前为止,简单的模型已经不能满足人们对预测精度的期望。本文提出的PCA-MLR-MEABP组合模型意在预测上证指数的走向,实现了两个结合:多种技术分析方法的结合、线性方法(主成分分析、多元线性回归)与非线性方法(思维进化算法优化的BP神经网络)的结合,具有很强的预测能力和实际应用价值。为了更好地提取上证指数的趋势信息,本文从股价、成交量、涨跌趋势、涨跌幅度等方面,选取BBI指标、MWVAD指标、乖离率(BIAS指标)、K值、D值、BOLL指标、WR指标、成交量八个技术指标,使用主成分分析法(PCA)进行指标的汇总和趋势分解;然后利用多元线性回归(MLR)的方法将主成分分析的结果整合成四个主成分变量;接着使用思维进化算法(MEA)优化BP神经网络的权值和阈值,最终构造出优化模型PCA-MLR-MEABP组合模型。为了说明组合模型的有效性,本文从拟合精度和趋势预测精度两个方面共选取了四个误差检验统计量,对PCA-MLR-MEABP组合模型、涉及到的PCA-MLR组合模型和BPNN模型、单一的神经网络模型:广义回归神经网络(GRNN)模型、交叉验证-随机森林(CV-RF)模型、Elman神经网络(ElmanNN)模型做误差检验,最终发现,PCA-MLR-MEABP组合模型在模型拟合度和趋势预测准确度方面都有显著优势。
[Abstract]:In the development of stock market, people have been working on how to accurately predict the stock price, at the same time, the forecasting methods are constantly updated. Up to now, the simple model can not meet the expectations of prediction accuracy. The PCA-MLR-MEABP combination model proposed in this paper is intended to predict the trend of the Shanghai Stock Exchange Index and realizes two combinations: the combination of various technical analysis methods, the linear method (principal component analysis, and the principal component analysis). The combination of multivariate linear regression and nonlinear method (BP neural network optimized by evolutionary algorithm) has strong predictive ability and practical application value. In order to better extract the trend information of the Shanghai Stock Exchange Index, this paper selects the BBI index from the aspects of stock price, trading volume, trend of rise and fall, the index of MWVAD, the rate of departure (BIAS), the index of K value and D value of bole index, the index of WR, and the eight technical indexes of trading volume. The principal component analysis (PCA) was used to summarize the index and the trend decomposition, and then the results of principal component analysis were integrated into four principal component variables by the method of multivariate linear regression (MLR). Then the weight and threshold of BP neural network are optimized by using the thought evolution algorithm (MEA), and the PCA-MLR-MEABP combinatorial model of the optimization model is constructed. In order to illustrate the validity of the combined model, this paper selects four error test statistics from two aspects of fitting accuracy and trend forecasting precision. For the PCA-MLR-MEABP combination model, the PCA-MLR combination model and the BPNN model are involved. Single neural network model: generalized regression neural network (GRNN) model, cross-validation-random forest CV-RFV model Elman neural network (Elman NNN) model for error testing, Finally, it is found that PCA-MLR-MEABP combination model has significant advantages in model fitting and trend prediction accuracy.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【参考文献】