A study on machine learning prediction of bio-oil yield from biomass and plastic Co-pyrolysis

Abstract

The co-pyrolysis of biomass and plastics can effectively enhance the quality of bio-oil. The application of machine learning techniques to predict bio-oil yield helps optimize the production of co-pyrolysis bio-oil. This study develops machine learning models for predicting bio-oil yield based on Deep Neural Networks (DNN) and Lightweight Gradient Boosting Machines. The study innovatively integrates the pyrolysis data of the three major components of biomass (cellulose, hemicellulose, and lignin), both individually and in mixtures, into the co-pyrolysis prediction model, overcoming the limitations of traditional studies that focus solely on the overall characteristics of biomass. The results show that the DNN model outperforms others, with the incorporation of biomass component data significantly improving the prediction accuracy of co-pyrolysis bio-oil yield, increasing the R² from 0.817 to 0.931, with an average absolute error of 3.583 and a root mean square error of 4.573. Additionally, analyses using Shapley additive explanations and Pearson correlation coefficients reveal significant changes in the feature importance ranking of the model, dynamically unveiling the impact mechanism of data expansion on feature weights. For the first time, the synergistic effect of plastic proportion and hydrogen content is explicitly identified. This research contributes to a deeper understanding of biomass pyrolysis mechanisms, thereby enhancing the economic value of co-pyrolysis bio-oil.