Boundary-aware data augmentation for solid waste hydrothermal carbonization: An uncertainty-guided SMOTE framework

Hydrothermal carbonization (HTC) plays a crucial role in sustainable organic solid waste conversion for mitigating energy crises. However, accurate prediction of HTC performance is challenged by the limited availability of experimental data, which restricts model performance. To overcome this limitation, a boundary-aware data augmentation framework integrating an uncertainty-guided synthetic minority over-sampling technique was developed. The framework incorporates an uncertainty quantification module to continuously adjust sampling weights, ensuring that synthetic samples are generated within a physically reliable parameter space. By integrating this mechanism with a boundary-adjusted sample generation approach, high-quality HTC samples can be produced. Furthermore, the framework is combined with gradient boosting regression to predict HTC performance, enabling accurate estimation of hydrochar HHV and yield. The case results showed that the proposed method achieved an R² of 0.890 and RMSE of 2.327 MJ/kg for hydrochar HHV prediction, and an R² of 0.831 and RMSE of 7.188% for hydrochar yield prediction. Compared with the model without data augmentation, the MAE and MAPE of hydrochar yield decreased by 14.33% and 13.71%, respectively, while the RMSE of hydrochar HHV decreased by 6.54%. These case results indicate that the proposed method effectively addresses the challenge of small-sample modeling for organic solid hydrothermal carbonization, providing a valuable reference.