Bio-oil production from biomass pyrolysis via a hybrid mathematical and machine learning approach

A hybrid mathematical and machine learning (HMML) approach is applied for maximizing bio-oil yield during the continuous biomass pyrolysis, using local common reed as feedstock. The HMML approach, which combines Response Surface Methodology (RSM) and Machine Learning algorithms (Gaussian Process (GPR) and Support Vector Regression (SVR)), provides a powerful tool for experimental optimization. Temperature, carrier gas flow rate, and feeding rate were the major parameters studied and optimized. The HMML approach demonstrated a strong predictive capability, as indicated by the statistical results. The optimal conditions (490 °C, 220 SCCM, 10.8 RPM) were identified and implemented, achieving a validated bio-oil yield of 39.90 ± 0.19 wt.%. The results confirm HMML as an effective tool for optimizing complex pyrolysis processes and improving bio-oil yield.