粒子群优化神经网络在多种股市中的预测研究

发布时间：2018-01-31 00:46

本文关键词： 股价预测 bp神经网络粒子群算法股票的可预测性　出处：《复旦大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着中国股票市场的发展,股票市场的投资活动逐渐变得频繁,股票市场逐渐成为证券市场中最活跃的市场,股票成为投资者们最热衷的投资产品。所以股票价格的预测成为了一项热门研究。有效的预测分析方法可以很好的帮助投资者制定投资策略,在增加收益的同时降低风险。股票市场是一个非常复杂的系统,但是它的内在规律具有一定的趋势性,而且受到经济政治等许多因素的影响,但就是如此股票市场这个系统的运动规律仍然是很难掌握的。许多的学者对股票市场进行研究,并且产生了许多的方法模型。传统的研究方法主要是基于数理统计理论的模型,先建立主观的数据序列模型,再对模型进行预测和研究,像时间序列等等模型在这方面有很多的应用,但是其预测精度无法达到人们的要求,事实上人们在研究中逐渐发现股票市场系统是一个复杂的非线性系统,传统的线性模型无法很好的逼近其内在规律,许多的学者开始研究股票的混沌性质,而且随着非线性算法的发展,许多学者开始使用神经网络,遗传算法等非线性算法对股票是场进行预测,各种基于非线性算法的股票预测模型被建立。本文将粒子群优化算法和bp神经网络进行了融合,利用粒子群算法对神经网络的连接权重和阈值的训练进行优化,讨论了各个参数的选取设定优化,建立了粒子群优化bp神经网络模型并将其用于股票预测的实证研究。通过对三种具有市场代表性的指数进行实证分析,以及将粒子群优化神经网络的预测效果与传统的单一bp神经网络预测效果进行对比分析,得到的结果表明粒子群算法能够有效的加强神经网络的预测能力,减小预测误差,提高训练速度；三大市场的预测结果都比较理想,说明了股票市场的可预测性；美国股票市场相对其他两种市场具有更强的预测性,规律性更强。
[Abstract]:With the development of China's stock market, the investment activities of the stock market become more and more frequent, and the stock market gradually becomes the most active market in the stock market. Stocks have become the most popular investment products for investors. Therefore, the prediction of stock prices has become a hot research. Effective forecasting and analysis methods can help investors to make investment strategies. Stock market is a very complex system, but its inherent law has certain tendency, and is influenced by many factors such as economy and politics. However, it is still very difficult to master the movement law of the stock market system. Many scholars study the stock market. Traditional research methods are mainly based on mathematical statistics theory model. First, the subjective data sequence model is established, then the model is predicted and studied. Models such as time series have many applications in this field, but their prediction accuracy can not meet the requirements of people. In fact, people have gradually found that the stock market system is a complex nonlinear system. The traditional linear model can not approach its inherent law very well, many scholars begin to study the chaos property of stock, and with the development of nonlinear algorithm, many scholars begin to use neural network. The nonlinear algorithm such as genetic algorithm is used to predict the stock field, and a variety of stock prediction models based on nonlinear algorithm are established. In this paper, particle swarm optimization algorithm and BP neural network are fused, the training of connection weight and threshold of neural network is optimized by particle swarm optimization, and the selection and optimization of each parameter are discussed. The BP neural network model of particle swarm optimization is established and applied to the empirical research of stock forecasting. Through the empirical analysis of three representative indices of market. The prediction effect of PSO neural network is compared with that of traditional BP neural network. The results show that PSO can effectively enhance the prediction ability of neural network. Reduce the prediction error and improve the training speed; The forecast results of the three major markets are all satisfactory, which shows the predictability of the stock market; The US stock market is more predictable and more regular than the other two markets.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP18;F832.51

【参考文献】