Use of machine learning tools to estimate drying rate of apple and apricot wood discs

This study compares three machine learning (ML) models -artificial neural networks (ANN), recurrent neural networks (RNN), and extreme gradient boosting (XGBoost)- for predicting the drying kinetics of wood biomass from the clearing of outdated apple and apricot orchards. The primary goal was to optimize and evaluate the performance of these models in predicting drying time and moisture content of the wood, which is crucial for improving energy efficiency in the drying process. Experiments were conducted under four temperature regimes (40°C, 50°C, 60°C, and 70°C) to simulate low- and high-temperature drying conditions. The models were trained and evaluated using statistical quality metrics such as root mean squared error (RMSE), mean absolute error MAE, and R². Results showed that all models performed with high accuracy, but XGBoost demonstrated the best statistical performance, exhibiting the lowest MAE and RMSE values. The RNN model exhibited the highest R² values and performed well in terms of MAE and RMSE, whereas the ANN model yielded slightly lower results. Despite small differences in performance, the models showed strong predictive capabilities and can be effectively used for modeling the drying process of moist wood biomass. This research emphasizes the significance of model optimization in enhancing the accuracy of drying time predictions and minimizing energy consumption in drying processes.