基于深度学习神经网络的季度GDP预测  

Quarterly GDP Forecasting Based on Deep Learning Neural Network

作　　者：梁龙跃 陈玉霞 Liang Longyue;Chen Yuxia(School of Economics,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学经济学院,贵阳550025

出　　处：《统计与决策》2023年第2期24-29,共6页Statistics & Decision

基　　金：贵州省省级科技计划项目(黔科合基础-ZK[2022]一般076);贵州大学人文社会科学一般课题(GDYB2022018,GDYB2022048);贵州大学经济学院研究生创新基金(CJ202071)。

摘　　要：针对季度GDP数据,文章基于深度学习长短期记忆(LSTM)神经网络模型,结合小波分析技术(WA)分解所选取的宏观经济变量,构建了LSTM&WA预测模型,同时,引入多个基准模型进行对比分析。研究表明:对于季度GDP数据,深度学习模型结合小波分析预测结果更优;针对结构复杂的非线性多变量数据,LSTM&WA预测模型具有较好的泛化能力,其预测精度均优于其他基准模型。For quarterly GDP data, this paper constructs an LSTM&WA forecasting model based on a deep learning long short-term memory(LSTM) neural network model combined with wavelet analysis(WA) technique to decompose the selected macroeconomic variables, while introducing several benchmark models for comparative analysis. The results show that for quarterly GDP data, the deep learning model combined with wavelet analysis has better forecasting results, that for nonlinear multivariate data with complex structure, LSTM&WA forecasting models have better generalization ability, and their forecasting accuracy is better than that of other benchmark models.

关 键 词：季度GDP预测 小波分析 深度学习 宏观经济变量 

分 类 号：F222[经济管理—国民经济] TP18[自动化与计算机技术—控制理论与控制工程]