摘要
提出了一种新的神经网络集成股市建模方法,采用偏最小二乘方法构造神经网络输入矩阵,利用Bagging技术和不同的神经网络学习算法生成集成个体,再用遗传算法选择参与集成的个体,以"误差绝对值和最小"为最优准,建立最小一乘回归神经网络集成模型,通过上证指数开盘价、收盘价进行实例分析,计算结果表明该方法具有较好的学习能力和泛化能力,在股市预测中预测精度高、稳定性好。
A novel neural network ensemble model is proposed for stock market forecasting. First of all, the partial least-square regression is used to extract input factors, and then many individual neural networks are generated by Bagging techniques and different training way. Secondly, genetic algorithm is used to select appropriate ensemble members. Finally, the least absolute regress method is used for neural network ensemble based on the least-absolute criteria. This method is established the forecast model of Shanghai stock exchange index. The result shows that the ensemble network has reinforcement learning capacities and generalization ability.
出处
《计算机工程与设计》
CSCD
北大核心
2007年第23期5812-5815,5818,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(40675023)
广西教育厅基金项目(200508234)
关键词
偏最小二乘回归
神经网络
遗传算法
神经网络集成
最小一乘回归
泛化能力
partial least squares regression
neural networks
genetic algorithms
neural network ensemble
least absolute regress
generalization ability
作者简介
吴建生（1974-），男，陕西咸阳人，硕士研究生，研究方向为神经网络及其自然进化算法应用。E-mail：wjsh2002168@163．com
