基于PCA的GA-BP网络对股票预测研究

发布时间：2018-06-13 17:55

本文选题：人工神经网络 + BP算法　；参考：《华东理工大学》2013年硕士论文

【摘要】：随着人们对投资思想的重视,人们在日常活动中越来越关注股市。然而股票投资属于一种高风险和高收益并存的投资领域,因此投资者们一直都非常关注有关股票价格的预测。自从股票市场开始出现,它就一直为国内外的许多学者所研究,同时众多的有关股票价格的预测方法也相应被提出。本文在基于各种分析之后提出了利用三层BP神经网络来构建股票预测模型。然而传统的BP网络尚存诸多不足之处,例如对初始权值的敏感、算法搜索时很难达到全局最优值、训练速率较慢等,因此应用于股票预测的效果欠佳。基于以上存在的缺陷,本文提出首先使用主成分分析法预处理网络输入变量,可以减少变量维数,降低股价数据的噪声。然后利用遗传算法优化网络参数,在网络训练过程中,选择LM算法以避免网络陷入局部极小值并促进网络的收敛速度。最后,详细讨论了网络的拓扑结构及其参数的确定原则,例如隐含层节点数和训练参数等。预测结果表明本文使用的优化算法的可行性。
[Abstract]:With the attention of people to the investment thought, people pay more and more attention to the stock market in their daily activities. However, stock investment is a high risk and high yield investment field, so investors have been very concerned about the stock price forecast. Since the emergence of stock market, it has been studied by many scholars both at home and abroad, and many forecasting methods about stock price have been put forward accordingly. In this paper, based on various analyses, a three-layer BP neural network is proposed to build stock forecasting model. However, the traditional BP network still has many shortcomings, such as sensitivity to initial weights, difficulty to reach the global optimal value in algorithm search, slow training rate, and so on, so the effect of applying it to stock prediction is not good. Based on the above defects, this paper proposes that the principal component analysis (PCA) is first used to preprocess the input variables of the network, which can reduce the dimension of variables and reduce the noise of stock price data. Then genetic algorithm is used to optimize the network parameters. In the process of network training, LM algorithm is selected to avoid the network falling into a local minimum and to promote the convergence speed of the network. Finally, the topological structure of the network and the determination principle of its parameters, such as the number of hidden layer nodes and the training parameters, are discussed in detail. The prediction results show the feasibility of the optimization algorithm used in this paper.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP183;F830.91

【参考文献】