Research on Pork Export Data Prediction Model Based on Temporal Neural Networks and Discrete GM Combination Model

This paper proposes a forecasting method based on a combination model of time series neural networks and discrete grey GM(1,1) to address the nonlinear, seasonal, and random characteristics present in pork export data. Our model can be divided into three modules: long-term trend modeling module, short-term fluctuation modeling module, and dynamic weight fusion module. The long-term trend modeling module employs an improved discrete grey GM(1,1) model to capture the overall trend of the data, thereby identifying the long-cycle variation patterns in pork export data. It transforms the original complex sequence into a more easily modeled monotonic sequence through cumulative generation and grey modeling processes, extracting the long-term variation patterns of the data. The short-term fluctuation modeling module introduces the time series neural network GRU, leveraging its advantages in handling temporal dependencies and nonlinear characteristics to perform secondary modeling on the prediction residuals of the grey model from the first module, capturing short-term fluctuations and complex nonlinear features, thus further extracting short-term fluctuations and seasonal characteristics. In the dynamic weight fusion module, a dynamic weight allocation mechanism is designed to adjust the contribution ratios of the two sub-models based on historical prediction errors, dynamically weighting and fusing the two prediction results to generate the final predicted value, thereby improving overall prediction accuracy. Experimental results show that the proposed combination model significantly reduces prediction errors on the pork export dataset compared to single models, demonstrating high performance and accuracy. This model provides an efficient and flexible solution for forecasting agricultural export trade data, with strong theoretical significance and practical application value.