Machine learning optimization for enhanced biomass-coal co-gasification

Abstract

The co-gasification of biomass feedstocks with coal offers a promising approach to enhancing syngas quality while mitigating the environmental impacts of traditional coal gasification. However, experimental determination of the optimal biomass/coal blending ratio and operational parameters is often resource-intensive. To address this challenge, modeling techniques are invaluable for optimizing biomass-coal co-gasification. This study aims to develop a machine learning (ML) model to optimize biomass-coal co-gasification. Additionally, an evolutionary algorithm is employed for multi-objective optimization, targeting maximum H2 production and optimal performance for the Fischer-Tropsch process. A comprehensive dataset from reputable literature sources, covering a wide range of biomass/coal blending ratios under various process conditions, was compiled. The dataset underwent statistical analysis, and mechanistic discussions were included to elucidate the effects of each parameter on the process. Among the four ML models applied, gradient boosting regression demonstrated the best performance during the testing phase, achieving an R2 exceeding 0.92 and MAE and RMSE values lower than 2.92 and 3.39, respectively. For H2 production, optimal results were observed with steam yields and temperatures near 1480 °C, while air and temperatures around 1570 °C yielded the best outcomes for the Fischer-Tropsch process. A biomass/coal blending ratio between 50 % and 70 % was found to be suitable for almost all gasifying agents under both criteria. The process was also analyzed techno-economically based on optimal conditions, revealing that steam exhibits superior techno-economic performance compared to other gasifying agents.