Biomass and polymer catalytic co-pyrolysis can convert waste into higher-quality fuels, thereby reducing the use of fossil fuels to some extent. However, this process is an extremely complex thermochemical conversion, influenced by numerous factors such as feedstock properties, operational variables, and catalyst. Currently, experimental methods require substantial time and resource investment. Machine learning (ML) can fit and match input and output features based on existing data, achieving extremely high accuracy in the co-pyrolysis process. This study applies advanced ML models to study the biomass and polymer catalytic co-pyrolysis process, with a focus on the yield of pyrolysis products and the variations of the oxygen-containing components in the pyrolysis oil. The best-performing model is used for feature analysis of the correlation between inputs and outputs, based on game theory SHAP analysis. The results indicate a significant negative correlation between the polymer addition ratio and the generation of oxygen-containing components during the co-pyrolysis process. The addition of catalysts promotes the generation of pyrolysis gas during co-pyrolysis but suppresses the yield of pyrolysis oil. Additionally, catalysts significantly inhibit the formation of oxygenates in the pyrolysis oil. The XGBR model shows the highest performance in predicting pyrolysis oil yield, achieving R2 values of 0.98 during training phase and 0.91 during testing phase. The GBR model performs well in predicting the oxygenate composition of pyrolysis oil from small datasets.
Read full abstract