基于神经网络和遗传算法的金融数据分析方法研究

发布时间：2018-01-13 17:43

本文关键词：基于神经网络和遗传算法的金融数据分析方法研究　出处：《广东财经大学》2017年硕士论文　论文类型：学位论文

【摘要】：股市是一个复杂的系统,股市研究是经济领域一个炙手可热的课题。神经网络具有良好的非线性系统拟合能力,然而采用神经网络模型分析预测股票,难以给出合适的变量选择准则。遗传算法基于达尔文“适者生存”理论,通过一个合适的适用度函数的“指导”,使得优质基因(优质个体)能够以较大的概率遗传给下一代得以存续。将这一方法用于变量选择,可以对影响股票价格的变量进行全局优化,有效地解决了神经网络输入层变量的选取问题。已有适用度函数只考虑预测误差,预测误差越小,个体的适用度越高。然而,当预测误差相同或相近时,我们更应该优先选择那些变量个数少的个体。显然,已有适用度函数并不能解决这一问题。基于这一思想,本文提出了一种新的适用度函数。新的适用度函数不仅考虑预测误差,同时也考虑变量的个数。基于新的适用度函数的“指导”得到的优质个体,既能保证好的预测结果,又具有较少的变量个数。本文选取沪深300指数数据,分别采用单一的人工神经网络模型(BPNN),主成分分析和神经网络组合模型(PCA-BPNN),遗传算法和神经网络组合模型(GA-BPNN)和本文改进的遗传算法和神经网络组合模型(IGA-BPNN)进行实验。结果表明,本文方法在保证基本相当的预测精度的同时,能有效减少变量个数。
[Abstract]:Stock market is a complex system, stock market research is a hot topic in the economic field. Neural network has a good nonlinear system fitting ability, but the neural network model is used to analyze and predict stocks. The genetic algorithm is based on Darwin's "survival of the fittest" theory, through the "guidance" of an appropriate applicability function. So that high-quality genes (high-quality individuals) can be inherited to the next generation with a high probability to survive. Using this method to select variables, we can optimize the variables that affect the stock price globally. The problem of selection of input layer variables in neural network is effectively solved. The prediction error is only considered in the existing applicability function. The smaller the prediction error, the higher the individual applicability. However, when the prediction error is the same or similar. We should give priority to those who have a small number of variables. Obviously, the existing function of degree of applicability can not solve this problem. In this paper, a new fitness function is proposed. The new fitness function not only considers the prediction error, but also considers the number of variables. This paper selects the index data of CSI 300 and uses a single artificial neural network model (BPNN). Principal component analysis (PCA) and neural network combined model (PCA-BPNN). The genetic algorithm and neural network combination model GA-BPNN) and the improved genetic algorithm and neural network combination model IGA-BPNN) are tested. This method can effectively reduce the number of variables while ensuring the basic equivalent prediction accuracy.
【学位授予单位】：广东财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F832.51

【参考文献】