Optimization model of agricultural economic structure based on LSTM

With the development of artificial intelligence (AI) and agricultural big data, the adjustment of agricultural economic structure has shifted from empirical judgment to data-driven decision-making. This study uses the provincial panel data of agricultural economy in Hubei Province from 2010 to 2023 to construct a comprehensive framework of prediction and optimization. The forecasting layer uses the Long Short-Term Memory (LSTM) model to capture the dynamic relationship among agricultural GDP growth rate, main crop output, rural employment structure, agricultural product price index, financial subsidies and agricultural infrastructure investment. The optimization layer uses the prediction results as parameters. It defines three objectives. overall agricultural benefit, structural deviation and food security risk. It also includes restrictions on cultivated land area, budget, labor supply and food security. Based on these backgrounds, the adjustment scheme of agricultural economic structure was formulated. The results show that the Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and r of LSTM model are 0.32, 0.45 and 0.91 respectively. Its prediction performance is better than autoregressive moving average (ARIMA), back propagation neural network (BPNN), support vector regression (SVR) and random forest (RF). These findings show that the combination of multivariate collaborative forecasting and structural optimization constraints can improve the efficiency of agricultural resource allocation. This method also provides quantitative support for rural revitalization, food security and digital agriculture policy design.