Prediction Of Lignocellulosic Biomass Pyrolysis Characteristics Based On Machine Learning

The extensive application of liquid fossil fuels not only leads to serious carbon emission problems,but also faces the depletion of traditional resources.As the only renewable organic carbon resources on earth,biomass can be converted into bio-oil,biochar and biogas through pyrolysis,which has great potential for development and application.In this work,based on machine learning algorithm,the prediction of lignocellulosic biomass pyrolysis process is studied.The research status of lignocellulosic biomass pyrolysis was investigated,and the research contents were divided into four aspects:pyrolysis kinetics,yield distribution of pyrolysis three-phase products,characteristics of bio-oil,and nitrogen gas release.Data sets were constructed on the pyrolysis kinetics,the yield distribution of three phase products（solid phase,liquid phase,and gas phase）,the characteristics of pyrolysis bio-oil（O/C,H/C,and heat value）,and the release of NH₃ and HCN in nitrogenous gas during pyrolysis of lignocellulosic biomass.Different machine learning models were constructed based on each data set,and their hyperparameters were optimized.Mean Square Error（MSE）and R² were used to evaluate the performance of each model.The ridge regression model,random forest model,and gradient boosted decision tree model were constructed by using the data set of pyrolysis kinetics.The optimal result was achieved by gradient boosted decision tree model with Bayesian optimization,the MSE was 198.67 and R² was 0.991.Based on the yield distribution of three-phase pyrolysis products（solid phase,liquid phase,and gas phase）,the characteristics of pyrolysis bio-oil（O/C,H/C,and heat value）,and the release of NH₃ and HCN during pyrolysis,random forest and gradient boosted decision tree models were constructed respectively.The optimal MSE/R² of the three-phase products（solid,liquid,and gas phases）were 13.11/0.978,14.06/0.948,and 14.16/0.952,respectively.The optimal MSE/R² of O/C,H/C,and heat value of pyrolysis bio-oil were 0.63/0.934,2.712/0.940,and 12.763/0.993,respectively.The optimal MSE/R² of the release of NH₃ and HCN during pyrolysis were 8.627/0.993 and 3.906/0.997,respectively.The growth process of the decision tree was visualized based on the gradient boosted decision tree model,the importance of features was analyzed,and the partial dependence analysis was carried out on the feature variables.The results showed that lignin content among biomass characteristics has the greatest influence on activation energy.When lignin content is 20-40 wt.%,C content is 45-60 wt.%,or fixed carbon content is 10-20 wt.%,the activation energy of biomass can be increased with the increase of various indexes.The experimental conditions had more influence on the pyrolysis products yields than the characteristics of biomass.Lignocellulosic biomass with high oxygen content,low carbon content,and high volatile content（>70 wt.%）could obtain more bio-char when pyrolysis at a lower temperature with lower heating rate.Lignocellulosic biomass with high ash content,high volatile content,high fixed carbon content,low oxygen content,high carbon content,and high wood content can obtain more bio-oil when pyrolysis at around 550°C.Lignocellulosic biomass with high oxygen content,high carbon content,high wood quality content,low ash content,and high volatile content（67 wt.%）could obtain more bio-gas during pyrolysis at high temperature.The pyrolysis temperature had the greatest influence on the characteristics of bio-oil.The bio-oil with high oxygen content often had low heat value.The high content of H element and low content of O element in biomass could improve the heat value of the bio-oil.The ratio of NH₃ and HCN release in nitrogen-containing gases gradually increased with the increasing pyrolysis temperature.The release amount of NH₃ could be significantly increased by increasing the temperature during the pyrolysis at low temperature,but the release of HCN could be significantly increased by increasing the temperature during pyrolysis at high temperature.Based on the above prediction results,on the one hand,it was proved that the model does"learn"the accurate law of the data set,and the model was reliable.On the other hand,it provided guidance for the selection of biomass materials and the control of pyrolysis conditions.