Hybrid framework combining machine learning with mechanism models for enhanced forecasting of biomass gasification products.

Owing to the inherent complexity and nonlinear variable interactions in biomass gasification, accurate products prediction with limited data poses a significant challenge. Herein, in this study, we proposed a hybrid model framework by mixing thermodynamic equilibrium models (Aspen Plus) with machine learning (ML) models. The improved simulator generative adversarial network (SimGAN) based on Wasserstein Loss was utilized to compensate the mismatch between the simulated and experimental data due to underlying assumptions. Subsequently, the artificial neural network (ANN) model showed the most excellent predictive performance with the average coefficient of determination (R2) and root mean square error (RMSE) of 0.931 and 1.146. Additionally, shapley additive explanations (SHAP) analysis and partial dependence plots (PDP) revealed the most influential features and the strong interactive effects on the gasification products. The proposed hybrid framework integrates data-driven with mechanism modeling approaches to improve the prediction accuracy of biomass gasification products.