【作者】 张迎春；

【导师】 赵远东；

【作者基本信息】 南京气象学院 ， 系统分析与集成， 2003， 硕士

【摘要】 本文提出了一个面向股市预测的模糊神经网络系统，并针对系统性能的改善进行了深入研究。 在对前馈神经网络的训练中，使用参数自适应方法实现了学习率、惯性因子的自我调节，以避免系统误差陷入局部最小，加快网络的收敛速度；提出了优化BP网络结构的实验研究方法，并给出了有关隐含层数和节点数选择以及再学习策略引进的研究结果。将该算法同传统BP算法的预测偏差平方和进行比较，结果证实网络的逼近精度及泛化能力均得到了极大的提高和改善。 我们采用上述优化算法，对深市某股票价格进行了基于模糊参量的神经网络模拟。为全面反映股市的特点和规律，该模型采集2001年3月1日至2003年4月11日500组原始数据，综合运用基本因素法和技术分析法，详细分析了影响股价走势预测效果的一些因素，提取出一定周期的回归性技术参数，以模糊时间序列的形式描述股价的变化趋势，并对未来相应周期内该股票价格上升的隶属度进行预测，最后进行准确度判决。理论分析及实验结果表明，该方法对股票市场的短期预测是可行和有效的；只要预测模型选取适当，即可获得超过市场平均盈利水平的收益。更多还原

【Abstract】 This paper puts forward a fuzzy neural network system aimed at stock price prediction, and an ameliorative method on its function is studied.In training of Back-Propagation neural network, parameter adaptable method which can automatically adjust learning rate and inertia factor is employed in order to avoiding systemic error immersed in a local minimum and accelerating the network’s convergence; Introduced the further optimization of the network’s structure, it gives the research result of selection of the hidden layers, neurons, and the strategy of re-learning, compared the sums of the deviation square of this algorithm with conventional BP algorithm, as a result, the approach accuracy and the generalization ability of the network were extremely improved.This improved BP Algorithm was here used to predict some stock prices of Shenzhen city to verify its accuracy and rationality, For the sake of roundly characterizing the trait and rule of the stock market, the model has gathered 500 groups of primary data from 3.1st, 2001 to 4.11th, 2003, labored some factors that influence the effect on predicting the trend of stock prices by synthetically applying the means of elementary factor and technique analysis, Thus we can extract the recursive technique parameter of proper-period, describe the vary trend of stock prices in form of fuzzy time series, and predict the ascending subjection degree of the stock prices in approaching homologous period and decide its accuracy. Academic analysis and experimental results show the method is feasible and effective on shortdated prediction of stock market; As long as a proper prediction model is chosen, people can get more profits than average market earning.更多还原