Soybean Futures Price Prediction Model Based on EEMD-NAGU

The soybean futures market in China occupies an important position in the agricultural product futures market. The research on the fluctuation of soybean futures price and the prediction of the future price trend has always been the focus of extensive attention in the field of agricultural economics. This paper proposes an EEMD-NAGU hybrid prediction model for soybean futures price based on Ensemble Empirical Mode Decomposition (EEMD) and New Attention Gate Unit (NAGU). The model first uses EEMD to process soybean futures price data into multiple Intrinsic Mode Functions (IMFs) and residual sequence. Then calculates the sample entropy of each IMF and reconstructed the IMF into three components of low-frequency, medium-frequency, and high-frequency, according to the size of the sample entropy. NAGU is formed by embedding the Attention mechanism (Attention) into the Gate Recurrent Unit (GRU) structure, further improving the learning capability of the model. NAGU splits the original reset gate and update gate, sets up two-stage respectively, and uses different activation functions to capture the information in historical data better. Soybean futures price data is complex nonlinearities and contain more noise. In this model, EEMD plays denoises the time series data and fixes the model input. NAGU can perform differential learning on data and finally produce prediction results. EEMD-NAGU is compared with thirteen other prediction models (Support Vector Regression (SVR), LSTM, GRU, NAGU, EEMD-LSTM, EEMD-GRU, EEMD-NGU, Attention-LSTM, Attention-GRU, Attention-NGU, EEMD-Attention-LSTM, EEMD-Attention-GRU, and EEMD-Attention-NGU). The evaluation indexes of the experiment are Mean Absolute Error (MAE), Mean Square Error (MSE), and R Squared ( $R^{2}$ ). The experimental results show that the EEMD-NAGU model outperforms other models with better prediction performance. The model can be widely used to predict the price of wheat, corn, gold, oil, and other time series data.