Abstract: Upright silos have been widely used as grain storage facilities, due to their high capacity and space efficiency. However, the uneven temperature distribution in silos can often cause potential grain spoilage and economic losses. Effective temperature management is required to reduce the post-harvest losses for high grain quality. Especially, there is an increasing global food demand and climate challenges. Since the existing research has focused on the temperature prediction models, it is still lacking in the necessary precision and adaptability, particularly in complex environments, like upright silos. This study aims to analyze the temperature distribution in a corn upright silo. A high-precision prediction model was then established for the real-time temperature control. A series of tests was conducted on a 5,000-ton corn upright silo in Wuhan, Hubei Province, China. The temperature data was collected from nine cables across 12 layers between July 2024 and January 2025. Pearson correlation analysis was implemented to assess the relationships between storage duration, internal/external temperatures and humidity, and temperature variations within the silo. The prediction models were constructed using Multiple linear regression (MLR), Lasso regression, Ridge regression, and XGBoost. The maximum and average temperatures were predicted to fully meet the temperature forecasts and gridded temperature predictions. Importantly, a hybrid model was proposed and then validated to combine with the Ridge regression and XGBoost. The results show that in the highest temperature prediction of layers 3 to 12, the Ridge regression model's average R²was 0.910 with an average MSE of 6.67, while XGBoost achieved an average R² of 0.942 and an average MSE of 8.15. The hybrid model demonstrated superior performance with an average R² of 0.944 and an average MSE of 3.75. In the average temperature prediction of layers 3 to 12, the Ridge regression model had an average R²of 0.904 and an average MSE of 5.86, whereas XGBoost achieved an average R² of 0.939 and an average MSE of 7.83. The hybrid model was further improved with an average R² of 0.946 and an average MSE of 3.63. Therefore, the hybrid model performed exceptionally well for the prediction of each temperature measurement point. There were the average R²values of 0.943 and 0.953 on the training and test sets, respectively, with the MSE values of 2.35 and 1.99, respectively. Overall, the highly accurate temperature prediction was achieved using Ridge regression and hybrid models. The great contribution was also gained for the real-time temperature control and grain storage optimization in the corn upright silos. Multiple environmental factors were integrated with the advanced statistical methods in order to enhance the model's robustness and generalization. The grid-based model can also provide the temperature predictions for each measurement point. Grain losses can be reduced after real-time temperature management in the grain storage facilities. Future research can further enhance the performance and adaptability in extreme conditions. Additional factors can be incorporated, such as the grain type, storage volume, and ventilation conditions. The research findings can offer valuable insights to improve the grain storage practices for food security. Additionally, the optimal models can be extended to the agricultural storage scenarios in food preservation. The models can be refined for the more specific data and real-time monitoring. The practical application can also offer a strong reference for the grain storage facilities in modern and intelligent agriculture.