林产化学与工业Issue(6)：31-41,11.DOI:10.20195/j.issn.0253-2417.2024191
基于机器学习的生物质热解炭产率预测的研究
Machine Learning-based Prediction for the Yield of Biomass Pyrolysis Char
摘要
Abstract
Four predictive models,including backpropagation neural network(BPNN),random forest(RF),support vector machine(SVM),and multiple linear regression(MLR),were established for biomass pyrolysis char yield,proximate analysis,and ultimate analysis.The performance of the models was evaluated using root mean square error(XRMSE).The results showed that the overall predictive accuracy ranked as BPNN=RF>SVM=MLR.Genetic algorithm and Bayesian optimization were applied to optimize the better-performing models.Furthermore,Pearson correlation coefficient(PCC),Shapley additive explanations(SHAP)values,and partial dependence plot(PDP)analysis were used to quantify the neural network,exploring the effects of biomass feedstock characteristics and pyrolysis conditions on the yield of char and product properties.The findings indicated that both genetic algorithm and Bayesian optimization could improve the generalization ability of the prediction model.The final pyrolysis temperature had the most significant effect on the yield of pyrolytic carbon,and the|PCC|value of the two was as high as 0.63.Moreover,machine learning revealed that the temperature-yield relationship was regulated by volatile content:biochar yield first increased and then decreased with temperature at low volatile content,while decreases continuously at high volatile content.A high mass ratio of C to H in biomass was beneficial for retaining C in the pyrolysis char.The fixed carbon and H in biomass collectively determined the final retention of oxygen in the pyrolysis char.关键词
生物质热解/热解炭产率/元素分析/工业分析/预测模型Key words
biomass pyrolysis/pyrolysis char yield/elemental analysis/industrial analysis/predictive modeling分类
化学化工引用本文复制引用
HUANG Zhangjun,XU Chenghui,CHEN Jie,DAI Jingchao,TIAN Hong..基于机器学习的生物质热解炭产率预测的研究[J].林产化学与工业,2025,(6):31-41,11.基金项目
国家自然科学基金面上项目(52476176) （52476176）
湖南省自然科学基金(2024JJ9179) （2024JJ9179）
