摘要
为提高神经网络经济预测的泛化能力,对神经网络预测的数据处理方法进行了改进,把对数据的归一化变为对数据增长率的归一化,因而只要预测的经济数据增长率不超过以往的经济数据增长率,则不再会发生外延问题.根据这一思路,采用北京市科技统计年鉴的数据,对经济发展进行了预测.预测结果与实际结果的比较说明改进有效.
To improve the generalization capability of the neural network, we optimize its application in processing data to forecast economics. In our work, normalization of data is changed into normalization of the increase rate from absolute value of economics. In this way, extensional phenomenon can be avoided as long as the predicted economic increase rate does not exceed those in the past. Based on this improved method, data of Beijing statistics were adopted in predicting the city＇s economics. The result proves that the method is gratifying and acceptable.
出处
《哈尔滨工业大学学报》
EI
CAS
CSCD
北大核心
2006年第6期897-898,916,共3页
Journal of Harbin Institute of Technology
基金
"十五"国家重点科技攻关项目(2003BA808A04)
关键词
神经网络
经济预测
泛化能力
MATLAB
neural network
economic forecast
generalization capability
MATLAB
