摘要
针对我国宏观经济的监测预测问题,提出了一种基于相关性分析的宏观经济指标预测算法。首先,通过插值和归类等方法对海量宏观经济指标进行预处理,获得频率一致的时序数据;然后,对预处理后的指标进行相关性分析,通过指标择优机制自动挑选出对指定的目标指标综合影响最大的关联指标集;最后,结合多元回归分析、反向传播(BP)神经网络等预测模型对目标指标进行预测,并计算每个关联指标对目标指标的贡献度。试验结果表明,该算法可有效提取关联指标,具有较强的鲁棒性和较高的预测准确率。
To address the problems of monitoring and forecasting China′s macro economy,a forecasting algorithm based on correlation analysis is proposed.Firstly,massive macroeconomic indicators are preprocessed by interpolation,classification and other methods to obtain the time-series data with consistent frequency.Then the correlation analysis of the preprocessed indicators is carried out,and the associated indicator set with the most comprehensive impact on the specific target indicator is automatically selected.Finally,the target indicator is predicted by the combination of multiple regression analysis,the back propagation(BP)neural network prediction model and other models,and the contribution of each associated indicator to the target indicator is calculated.The experimental results show that this algorithm,with strong robustness and high accuracy of prediction,can select associated indicators efficiently.
作者
李野
鲁淑
袁翔
袁林
LI Ye;LU Shu;YUAN Xiang;YUAN Lin(The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210007,China)
出处
《指挥信息系统与技术》
2020年第1期84-88,100,共6页
Command Information System and Technology
基金
国家重点研发计划课题(2018YFC0806902)资助项目。
关键词
宏观经济指标
相关性分析
多元回归分析
BP神经网络
预测
macroeconomic indicators
correlation analysis
multiple regression analysis
back propagation(BP)neural network
prediction