Co–pyrolysis of de–alkalized lignin and coconut shell via TG/DTG–FTIR and machine learning methods: pyrolysis characteristics, gas products, and thermo–kinetics. (1st December 2022)

• Record Type: Journal Article
• Authors: Yin, Hongchao; Huang, Xiankun; Song, Xiaohan; Miao, Hongchao; Mu, Lin
• Is Part Of: Fuel. Volume 329(2022)
• Publication Date: 2022-12-01
• Subjects: Lignin waste -- Co–pyrolysis -- Thermo–kinetics -- BP–NN -- Stacked XGB–LGBM–MLP model; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2022.125517
• Languages: English
• ISSNs: 0016-2361

Abstract

Graphical abstract: Highlights: Application of machine learning methods to DL–CS co–pyrolysis. D10C90 inhibits CO2 production, which has implications for carbon neutrality. Gaseous functional groups during the pyrolysis were examined by TG–FTIR. Estimation of activation energy by KAS, FWO, and Starink methods. The stacked XGB–LGBM–MLP model preform best of all individual models. Abstract: Co–pyrolysis characteristics and thermo–kinetics of de–alkalized lignin (DL) and coconut shell (CS) were investigated by using TG–FTIR and machine learning methods. The DL–CS samples mixed in different proportions were heated from ambient temperature to 1173.15 K at 10, 15, and 20 K·min −1 . The gas functional groups in the pyrolysis process of the experimental samples were detected by TG–FTIR. The apparent activation energy ( E ) was estimated by Flynne–Walle–Ozawa (FWO), Kissinger–Akahira–Sunose (KAS), and Starink methods, and the R 2 values of these three methods are greater than 0.97032, which means that the activation energy solved by these methods is feasible. An BP–NN model of 9×3×1 architecture was employed to predict the residual weight of DL–CS co–pyrolysis. The experimental values were in good agreement with the predicted values by BP–NN model (RMSE = 0.8606, R 2 = 0.99888). A Stacked XGB–LGBM–MLP model was employed to improve the overall prediction performance of DL–CS co–pyrolysis, and this model succeeds to preform best (RMSE = 0.1888, R 2 = 0.99995) of all individual models from test set result parameters. Our research results contribute to the optimal operating conditions for energy utilization, pollution control, and thermochemical conversion of lignin waste industry.