Prediction of corn futures prices with decomposition and hybrid deep learning models

Predicting corn futures prices is crucial for market participants, policymakers, and agricultural enterprises to manage risks and make informed decisions. Therefore, it is necessary to improve the accuracy and stability of corn futures price predictions. This study proposes a deep mixed model that integrates variational mode decomposition (VMD), particle swarm optimization (PSO), convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and an attention mechanism. First, VMD is used to decompose the corn futures closing price series, thereby reducing data complexity. Then, principal component analysis (PCA) is performed on 21 influencing factors to extract key components affecting price fluctuations. The proposed model combines CNN for local feature extraction, LSTM for time series processing, and an attention mechanism to focus on critical information. In addition, the PSO algorithm optimizes the LSTM hyperparameters, further enhancing the model’s performance. To verify the model’s effectiveness, comprehensive experiments were conducted using historical data on corn futures. The results of four comparative experiments show that, based on multiple metrics such as RMSE, MAE, and \(\text {R}^2\) ${\text{R}}^2$ , the proposed model significantly outperforms 14 existing models in terms of prediction accuracy and stability. These findings indicate that the model has practical application potential in financial forecasting and decision-making.