Dynamic Prediction and Policy Simulation of Macroeconomic Indicators Based on Deep Learning Algorithm

As the macroeconomic operation mechanism becomes increasingly complex, traditional economic forecasting models face problems such as low prediction accuracy and slow dynamic response when dealing with high-dimensional, multivariable, and nonlinear economic systems. To this end, this paper first constructs a multivariable input system, adds exogenous variables such as international oil prices and policy factors, and enhances the time series characteristics through difference, normalization, and lag structure design. Then, based on the GRU model, an economic multi-step forecasting framework is constructed, and “policy scenario variables” are injected into the model input structure to simulate the future economic evolution path under the policy intervention scenario. The experimental results show that in the policy scenario simulation, the model can effectively predict the changing trend of the annualized GDP growth rate under loose and tight policies, providing a quantitative basis with timeliness and foresight for macroeconomic regulation.