一种改进的IPSO-BP神经网络在股指预测中的应用——以上证综指为例

An improved IPSO-BP neural network in stock market index forecasting——A case study of Shanghai Composite Index

针对股票数据具有规模庞大、结构复杂、多噪声和高度模糊非线性等特点而导致预测难的问题,利用改进的粒子群算法(固定惯性因子动态化)优化BP网络权阀值,建立了一个基于历史日收盘价、最低价、最高价、成交量、成交额、涨跌幅为输入变量,日开盘价为输出变量的预测模型.利用MATLAB软件对2007年1月4日至2015年8月31日上证综指(开盘价)进行了仿真预测,并且从绝对误差与相对误差等角度对比分析了BP网络优化前后的预测结果,结果表明IPSO优化后的BP网络不仅可以更快地实现收敛寻优,而且在对未来股价的趋势判断与指数预测方面均具有较好的预测效果.

For the stock data has the characteristics of large scale,complex structure,multi-noise and highly fuzzy nonlinearity,resulting in the forecast for the stock index has been a difficult problem.This paper uses an Back Propagation Neural Network Optimized by an Improved Particle Swarm algorithm（Dynamic the Fixed Inertia Factor）to establishe a forecasting model based on the historical closing price,the lowest price,the highest price,the trading volume,the turnover,the rising price as the input variable and the daily opening price as the output variable.Using MATLAB software to predict the Shanghai Composite Index（opening price）from Jan.4,2007 to Aug.31,2015,and the results of BP Neural Network Optimized before and after was compared and analyzed in terms of absolute error and relative error.It is concluded that the BP Neural Network Optimized by the IPSO algorithm has a better effect on the future stock price trend judgment and index forecasting.