Machine Learning-Based Prediction of Gaseous Fuel Generation During Co-Pyrolysis of Lignin and Red Mud

초록

This study systematically investigated the generation behavior of gaseous fuels (e.g., H2, CO, and CH4) during the pyrolysis of waste materials (lignin and red mud) and developed machine learning (ML)-based models for accurate prediction of gas yields. Pyrolysis experiments were carried out under varying reaction atmospheres (i.e., N2, CO2, and N2/CO2) and waste material mixing ratios to comprehensively evaluate the gas generation characteristics. Pearson correlation analysis identified reaction temperature and gas flow rate as the most influential parameters governing gas yields. Three ML models—artificial neural network (ANN), XGBoost (XGB), and Extra Trees (ET)—were developed to compare the gas generation prediction performances. The ANN model achieved superior accuracy, with coefficients of determination of R2 = 0.992 (H2), 0.989 (CH4), and 0.994 (CO). SHapley Additive exPlanations (SHAP) analysis was subsequently applied to quantify the relative contribution of each input variable to model predictions. Building on the ANN model, two-dimensional (2D) simulations were performed to explore the combined effects of temperature, gas flow rate, and red mud content on gas generation. Simulation results revealed consistent increases in H2, CO, and CH4 yields under specific combinations of these conditions. Maximum predicted yields of H2 (>3.5 mol%), CH4 (>1.2 mol%), and CO (>7.0 mol%) were identified under their respective optimal conditions. These findings demonstrate that ML-based approaches offer a robust and interpretable framework for predicting gas generation behavior and optimizing process conditions in pyrolysis systems.

키워드