A Time Series Decomposition-Driven Combined Model for Agricultural Price Forecasting

Accurate forecasting of agricultural product prices is vital for stabilizing market dynamics and informing effective policy decisions. However, the strong nonlinearity and complexity of influencing factors often limit the performance of traditional models. To address these challenges, this study proposes a hybrid forecasting framework—GRU-IHEOA-N-BEATS-BP—targeted at apple and crucian carp price prediction. The model first utilizes STL decomposition to separate the time series into seasonal, trend, and residual components. Pearson correlation analysis is applied to identify key influencing factors for each component. A GRU network is then constructed to model sequential dependencies, with its parameters optimized by an improved Human Evolutionary Optimization Algorithm (IHEOA) to enhance prediction accuracy. Concurrently, an N-BEATS model captures high-order nonlinear patterns from the decomposed components. Finally, a BP neural network aggregates outputs from all forecasting submodels to generate the final prediction. Experimental results validate the effectiveness of the proposed model. For crucian carp, the value R ² improves from 64. 67% (with N-BEATS standalone) to 97. 95\%, while MAE and MSE decrease by 81.5% and 93.6%, respectively. These results demonstrate the model’s superior ability to capture complex temporal and nonlinear characteristics. In summary, this hybrid approach significantly enhances the accuracy and robustness of forecasting. It provides a valuable tool for intelligent agricultural management, contributing to food security, higher farmer income, and more precise market regulation.