Data-driven modeling of biomass gasification and multi-criteria performance assessment for integrated power generation

A restricted, equilibrium-based, Aspen Plus model is developed for biomass gasification followed by gas compression for power generation. The gasifier includes explicit stoichiometric modeling of methanation to enable accurate CH4 yield prediction, overcoming this critical issue in available models. Validation with experimental data is conducted with both steam and air. We vary the six inputs of equivalence ratio, gasification zone temperature and pressure, gas and steam compressor outlet pressure and steam flowrate. The seven outputs include the syngas lower heating value, the cold gas efficiency, the thermal and carbon conversion efficiency, the total device duty, the syngas yield and the CO2 mole fraction. The resulting 54,540 feasible operating points are used to train eight algorithms including Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Random Forest Regressor (RF), Extreme Gradient Boosting (XGBoost), CatBoost, Light Gradient Boosting Machine (LightGBM), K-Nearest Neighbors (KNN), and Extra Trees Regressor (ETR). The predictions are compared with the Aspen model data, whereas multi-criteria assessment is implemented to compare their Pareto fronts. The proposed Aspen model indicates improvements in the accuracy of CO, H2, CH4 and CO2 concentrations compared to existing models. RF and ETR exhibit the highest predictive accuracy and excellent matching of the Aspen Pareto fronts.