人工神经网络.docVIP

摘要 随着经济的发展和人们投资意识的转变，股票投资已成为现代人 生活中一个重要组成部分，而股票价格的预测也成为投资者关心和研 究的重点。由于股票投资的收益与风险往往是成正比的，如何建立一 个运算速度和精确度都比较高的股市预测模型，对于金融投资者具有 理论意义和实际应用价值。 本文在深入分析股票市场预测面临的关键问题和比较各种股票 预测方法的基础上，探讨利用BP(Back Propagation)神经网络对股 票走势进行分析和预测的可行性。BP网络通过对以往历史数据的学 习，找出股市发展的内在规律，并将其存储在网络具体的权值、阀值 中，用以预测未来的走势。 针对BP算法在股市预测中存在的学习速度慢、容易陷入局部极 小值、预测结果精度不高等问题，提出一种改进的BP神经网络算法。 通过重新选取神经元的激活函数，对输出层和隐层中神经元转换函数 的权值、缩放系数和位移参数进行调整，减少隐层节点数，加快BP 网络的收敛速度。根据BP网络进行股市预测的原理，建立基于BP网络的股市预测模型，采用改进后的BP算法进行股市预测，并通过MATLAB软件对其预测过程进行仿真实验。 关键词 神经网络，股票预测，BP算法，激活函数 ABSTRACT With the economic growth and the conversion of people’S investment consciousness，stock has become an important part of people’Slife in modem time．Stock forecast has greatly been one of focuses of public topic．The proceeds of stock investment always equal the risk．So establishing a stock forecasting model，which has higher operation rate and precision，has theoretical significance and applicable value． This dissertation analyses the main problems being existent in the process of stock market prediction and compares various stock forecasting methods．The feasibility of forecasting stock trend by using BP neural network iS discussed．BP neural network finds out the disciplinarian of stock market through learning historical datum and store them in the weights and valve values of the neural network forforecasting the trend in the future． Based on studying these existing problems of BP algorithms in stack forecasting，including the slow leaming speed，local extremum and the low prediction precision，an improved BP neural network algorithm iS presented．This algorithm reduces number of hidden nodes and enhances the convergence speed by re-selecting activation function and adjusting weight value of transformation function，scaling coefficient and displacement parameter in output layer and hidden layer．According to the principle of stock prediction based on BP network，the prediction model of stock has been established．The stock iS predicted by adopting