Prediction and influencing factors analysis of stored grain temperature and intergranular relative humidity

Abstract

To address the issues of insufficient accuracy and poor interpretability in the prediction of temperature and intergranular relative humidity (RH) for bulk-stored grain, this study developed an optimized prediction model based on the extreme gradient boosting (XGBoost) algorithm. The model incorporated the interaction of meteorological factors with the temperature and RH within the bulk grain. Hyperparameters were optimized using the grid search method, and Shapley Additive exPlanations (SHAP) was employed to quantify the contribution and interaction mechanism of each input variable, thereby enhancing model interpretability. Seven meteorological variables and the surface RH of the bulk grain were used to predict the surface temperature. Similarly, meteorological data and surface temperature were used to predict the surface RH, and the model performance was validated using an experimental silo dataset. For benchmarking purposes, several commonly used models, including Linear Regression (LR), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and Support Vector Regression (SVR) were also tested. The results indicated that the XGBoost model outperformed the baseline models in predictive accuracy. The RMSE, R², MAE, and MAPE results for average surface temperature prediction were 1.59, 0.88, 0.79, 3.56 %, respectively, while those for average surface RH were 0.32, 0.95, 0.16, 0.28 %. SHAP analysis revealed that the influence of meteorological variables on grain surface temperature and intergranular RH was complex, with air temperature, air pressure, and air RH identified as the most significant factors. The proposed model demonstrated both high predictive accuracy and strong interpretability, offering a scientific basis for the monitoring and control of grain storage conditions.