Construction and Performance Evaluation of Grain Porosity Prediction Models Based on Metaheuristic Algorithms and Machine Learning

Grain porosity is a key parameter affecting the heat and moisture transfer in grain piles, and is related to the accuracy of heat and moisture transfer calculations and simulation studies in grain piles. To effectively measure the porosity of grains under different pressures, this study used a powerful machine learning algorithm to set six input parameters affecting porosity and experimentally acquired the data required for machine learning modeling. Eight machine learning-based models for grain porosity prediction were developed. Four meta-heuristic algorithms (i.e., Particle Swarm Optimization (PSO), Sine Cosine Algorithm (SCA), Harris Hawks Optimization (HHO), and Multi-Verse Optimizer (MVO)) were combined with Support Vector Regression (SVR) to obtain four hybrid models (i.e., PSO-SVR, SCA-SVR, HHO-SVR, and MVO-SVR), respectively, in order to improve the accuracy and methodology of prediction. In addition, four other intelligent models, namely SVR, least-squares support vector regression, random forest, and partial least-squares regression, were constructed and compared with the four hybrid models. Finally, the accuracies of the eight models were evaluated and compared using several metrics: coefficient of determination (R²), variance accounted for (VAF), Nash-Sutcliffe efficiency (NSE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and scatter index (SCI). The results indicate that the HHO-SVR model achieved the highest accuracy during both the training and testing phases, with R² values of 0.97660 and 0.95613, respectively. The HHO-SVR model predicts grain porosity with reasonable accuracy and exhibits superior performance compared to other models.