基于BP神经网络的股票价格预测输入变量选择研究

发布时间：2018-08-07 07:50

【摘要】：股票市场是一个高度复杂的非线性系统。股市的变化既有其自身规律性，又受政治、经济、投资心理等诸多因素的影响。传统的基于数理统计的预测方法很难对其进行有效地描述，而具备解决非线性问题能力、网络学习能力和系统拟合能力的人工神经网络可以在任意精度内实现变量间的非线性关系的映像，逼近证券价格随时间变换的函数，从而对股票市场进行模拟和学习。迄今为止，针对不同的股市，国外许多学者都建立了很多相应的预测模型，给出了很好的预测方法，也取得了良好的预测效果。但由于我国证券市场仅有二十多年的发展历史，还很不完善，，国外成熟市场上流行和行之有效的经验和方法未必适合目前中国股票市场。BP神经网络是一种常用股票价格预测方法，它具有强大的非线性拟合能力，许多学者在这一领域进行了深入的研究。但由于股票市场可选用预测参数太多，使BP神经网络内部运算混乱，常常导致运算量过大，而且精确度下降。因此，本文在国内外研究基础上，提出了一种股票价格预测的BP神经网络输入变量选择方法。首先采用主成分分析法降低输入向量的维数；然后采用层次分析法和德尔菲法相结合的方法调整输入向量的信息结构；最后将2种方法得到的输入向量组进行了仿真实验进行比较。结果表明，综合主成分分析法和结合层次分析法的德尔菲法得到的改进主成分向量组对于BP神经网络股票预测具有较好的性能。
[Abstract]:The stock market is a highly complex nonlinear system. The change of stock market has its own regularity, and it is influenced by many factors such as politics, economy, investment psychology and so on. The traditional method of forecasting based on mathematical statistics is difficult to describe it effectively, but has the ability to solve non linear problems, network learning ability and system fitting. The artificial neural network of force can realize the image of the nonlinear relation between variables in any precision, and approximate the function of the change of the stock price with time, so as to simulate and learn the stock market.
So far, many foreign scholars have set up a number of corresponding prediction models for different stock markets, give a good prediction method, and have achieved good prediction results. However, because China's securities market has only more than 20 years of development history, it is still very imperfect, the popular and effective foreign market experience and methods are not. The.BP neural network in Chinese stock market is a kind of common stock price prediction method, which has strong non-linear fitting ability. Many scholars have carried on deep research in this field. However, because the stock market can choose too many predictive parameters to make the internal operation of BP neural network chaotic, and often leads to too much computation. Therefore, on the basis of research at home and abroad, this paper puts forward a BP neural network input variable selection method for stock price prediction. Firstly, the principal component analysis method is used to reduce the dimension of the input vector. Then the information structure of the input vector is adjusted by the combination of AHP and Delphi method; finally, 2 The simulation experiment of the input vector group obtained by the method is compared. The results show that the improved principal component vector group obtained by the integrated principal component analysis method and the analytic hierarchy process (AHP) method by Delphy Fa has good performance for the stock prediction of BP neural network.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F832.51

【参考文献】