基于图注意力LSTM深度学习的季度GDP预测应用  

Quarterly GDP Forecasting Application Based on Graph Attention LSTM Deep Learning

作　　者：龙志 陈湘州[1,2,3] LONG Zhi;CHEN Xiangzhou(Business School,Hunan University of Science and Technology,Xiangtan 411201,China;Hunan Provincial Strategic Emerging Industry Research Base,Xiangtan 411201,China;Hunan New Industrialization Research Base,Xiangtan 411201,China)

机构地区：[1]湖南科技大学商学院,湖南湘潭411201 [2]湖南省战略性新兴产业研究基地,湖南湘潭411201 [3]湖南省新型工业化研究基地,湖南湘潭411201

出　　处：《湖南工程学院学报（社会科学版）》2024年第1期54-64,118,共12页Journal of Hunan Institute of Engineering(Social Science Edition)

基　　金：国家社会科学基金资助项目“持续调控背景下房地产市场利益分配协调机制及政策研究”(13BJY057)。

摘　　要：GDP预测一直以来都是宏观经济领域的重要研究议题。同时,融合深度学习算法并实时监测GDP的动态变化已是宏观经济指标预测的必然趋势。基于此,本文考虑到宏观经济变量的多源非线性时空特征,提出了一种新的GAT LSTM融合深度学习模型并对季度GDP进行预测。该模型采用图注意力网络(GAT)捕捉空间拓扑结构信息,并运用长短期记忆神经网络(LSTM)提取时间序列信息,以提高模型预测效果。结果表明:与其他基准模型相比,本文所提出的GAT LSTM模型在回归拟合度(R^(2))上平均提升了0.5929、均方误差(MSE)上平均下降了0.6617,具有良好的泛化能力和鲁棒性。该模型在GDP预测领域具有较好的应用前景,以帮助投资者、企业和国家作出科学决策,提高经济效益。GDP forecasting has always been an important research topic in the field of macroeconomics.At the same time,integrating deep learning algorithms and monitoring the dynamic changes of GDP in real time has been an inevitable trend in macroeconomic indicator forecasting.Based on this,this paper proposes a new GAT LSTM fusion deep learning model and forecasts quarterly GDP considering the multi source nonlinear spatio temporal characteristics of macroeconomic variables.The model uses a graph attention network(GAT)to capture spatial topological structure information and a long short term memory neural network(LSTM)to extract time series information to improve the model prediction effect.The results show that compared with the rest of the benchmark models,the GAT LSTM model proposed in this paper has good generalization ability and robustness,with an average improvement of 0.5929 in regression fit(R^(2))and a decrease of 0.6617 in mean square error(MSE).The model has a good application prospect in the field of GDP forecasting for investors,enterprises,and the state to make scientific decisions and improve economic efficiency.

关 键 词：GDP预测 图注意力网络 长短期记忆神经网络 深度学习 

分 类 号：F124[经济管理—世界经济] TP183[自动化与计算机技术—控制理论与控制工程]