Please wait a minute...
浙江大学学报(工学版)
Journal of ZheJiang University (Engineering Science)  2019, Vol. 53 Issue (3): 605-612    DOI: 10.3785/j.issn.1008-973X.2019.03.023
Chemical Engineering, Energy Engineering     
Random forest model for predicting kinetic parameters of biomass devolatilization
Jiang-kuan XING(),Hai-ou WANG,Kun LUO*(),Yun BAI,Jian-ren FAN
State Key Laboratory of Clean Energy Utilization, Zhejiang University, Hangzhou 310027, China
Download: HTML     PDF(1164KB) HTML
Export: BibTeX | EndNote (RIS)      

Abstract  

Two biomass devolatilization databases, including the training and validation databases, were constructed from diverse available experiments in literature. The kinetic parameters were fitted under the framework of order-based global biomass devolatilization model. The random forest (RF) method was employed to investigate the complex nonlinear correlations between the kinetic parameters with chemical compositions and heating condition, in order to develop the RF model to accurately predict the kinetic parameters of biomass devolatilization based on its chemical compositions and heating condition. The training and validation results show that the RF model can well predict the kinetic parameters of different biomass types under different heating rates (determination coefficient R2>0.92), also with an accurate prediction of the biomass devolatilization process (R2>0.93). The variable importance measurement (VIM) results show that the fraction of cellulose (CL) has significant effect on the reaction order and activation energy, and the fraction of lignin (LIG) has the maximum effect on the reaction order. The effect of heating rate is negligible for activation energy, but considerable for the frequency factor and reaction order.

Key words： biomass devolatilization      random forest (RF)      kinetic parameters      chemical composition      heating rate     
Received: 16 August 2018      Published: 04 March 2019
CLC:  TK 6  
Corresponding Authors: Kun LUO     E-mail: zjuxjk@zju.edu.cn;zjulk@zju.edu.cn
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Jiang-kuan XING
Hai-ou WANG
Kun LUO
Yun BAI
Jian-ren FAN
Cite this article:

Jiang-kuan XING,Hai-ou WANG,Kun LUO,Yun BAI,Jian-ren FAN. Random forest model for predicting kinetic parameters of biomass devolatilization. Journal of ZheJiang University (Engineering Science), 2019, 53(3): 605-612.

URL:

http://www.zjujournals.com/eng/10.3785/j.issn.1008-973X.2019.03.023     OR     http://www.zjujournals.com/eng/Y2019/V53/I3/605


预测生物质热解动力学参数的随机森林模型

基于大量已发表的生物质热解实验数据，采用数值方法拟合全局反应热解模型的动力学参数，建立生物质热解的训练和验证数据库，并利用随机森林算法研究生物质热解动力学参数与生物质种类和加热条件之间的非线性关系，发展预测生物质热解动力学参数的随机森林模型. 训练和验证的结果显示：随机森林模型能够较好地预测训练数据库中的生物质热解的动力学参数(R2>0.92)，并能够准确预测验证数据库中的多种生物质的热解过程(R2>0.93). 此外，变量重要性分析结果显示：纤维素质量分数对于反应级数和活化能影响较大，木质素对于反应级数的影响最大. 加热条件对于活化能的影响可以忽略，但是对指前因子和反应级数的影响显著.


关键词： 生物质热解,  随机森林(RF),  动力学参数,  化学组成,  升温速率 
β/（K·min?1） K/s?1 E/（J·mol-1） n
10 1.606×107 9.794×104 2.089
40 5.329×108 1.111×105 2.294
Tab.1 Fitted kinetic parameters for devolatilization process of eucalyptus leaves under different heating rates
Fig.1 Comparisons of devolatilization processes predicted by fitted kinetic parameters and experimental data for Eucalyptus leaves biomass
Fig.2 Schematic diagram of procedures for random forest (RF) method
Fig.3 Data distribution in training and validation database
Fig.4 Test results for determining optimal tree number in RF model
Fig.5 Training results of RF model for all three kinetic parameters (solid line in each subgraph means best performance)
Fig.6 Validation results of RF model for all three kinetic parameters (solid line in each subgraph represents best performance)
Fig.7 Comparisons of devolatilization process predicted through kinetic parameters from RF model and experiments for eucalyptus bark and eucalyptus sawdust under different heating rates
Fig.8 Measured relative importance of each input parameter on each kinetic parameter
[1]   GOYAL H B, SEAL D, SAXENA R C Bio-fuels from thermochemical conversion of renewable resources: a review[J]. Renewable and Sustainable Energy Reviews, 2008, 12 (2): 504- 517
doi: 10.1016/j.rser.2006.07.014
[2]   CAI J M, LIU R H New distributed activation energy model: numerical solution and application to pyrolysis kinetics of some types of biomass[J]. Bioresource Technology, 2008, 99 (8): 2795- 2799
doi: 10.1016/j.biortech.2007.06.033
[3]   SHENG C, AZEVEDO J Modeling biomass devolatilization using the chemical percolation devolatilization model for the main components[J]. Proceedings of the Combustion Institute, 2002, 29 (1): 407- 414
doi: 10.1016/S1540-7489(02)80054-2
[4]   BESTE A, BUCHANAN A C Kinetic simulation of the thermal degradation of phenethyl phenyl ether, a model compound for the β-O-4 linkage in lignin [J]. Chemical Physics Letters, 2012, 550 (8): 19- 24
[5]   COUHERT C, COMMANDRE J M, SALVADOR S Is it possible to predict gas yields of any biomass after rapid pyrolysis at high temperature from its composition in cellulose, hemicellulose and lignin?[J]. Fuel, 2009, 88 (3): 408- 417
doi: 10.1016/j.fuel.2008.09.019
[6]   PETERS B Prediction of pyrolysis of pistachio shells based on its components hemicellulose, cellulose and lignin[J]. Fuel Processing Technology, 2011, 92 (10): 1993- 1998
doi: 10.1016/j.fuproc.2011.05.023
[7]   KIM S W Prediction of product distribution in fine biomass pyrolysis in fluidized beds based on proximate analysis[J]. Bioresource Technology, 2015, 175: 275- 283
doi: 10.1016/j.biortech.2014.10.107
[8]   LECUN Y, BOSER B E, DENKER J S Backpropagation applied to handwritten zip code recognition[J]. Neural Computation, 1989, 1 (4): 541- 551
doi: 10.1162/neco.1989.1.4.541
[9]   HO T K. Random decision forests [C] // Proceedings of the 3rd ICDAR. Montreal: IEEE, 1995: 278–282.
[10]   XING J K, LUO K, HEINZ P, et al Predicting kinetic parameters for coal devolatilization by means of artificial neural networks[J]. Proceedings of the Combustion Institute, 2018,
[11]   LUO K, XING J K, BAI Y, et al Prediction of product distribution in coal devolatilization by an artificial neural network model[J]. Combustion and Flame, 2018, 193: 283- 294
doi: 10.1016/j.combustflame.2018.03.016
[12]   LEI C K A random forest approach for predicting coal spontaneous combustion[J]. Fuel, 2018, 223: 63- 73
doi: 10.1016/j.fuel.2018.03.005
[13]   SUNPHORKA S, CHALERMSINSUWAN B, PIUMSOMBOON P Artificial neural network model for the prediction of kinetic parameters of biomass pyrolysis from its constituents[J]. Fuel, 2017, 193: 142- 158
doi: 10.1016/j.fuel.2016.12.046
[14]   SLOPIECKA K, BARTOCCI P, FANTOZZI F Thermogravimetric analysis and kinetic study of poplar wood pyrolysis[J]. Applied Energy, 2012, 97: 491- 497
doi: 10.1016/j.apenergy.2011.12.056
[15]   XU T T, XU F, HU Z Q, et al Non-isothermal kinetics of biomass-pyrolysis-derived-tar (BPDT) thermal decomposition via thermogravimetric analysis[J]. Energy Conversion and Management, 2017, 138: 452- 460
doi: 10.1016/j.enconman.2017.02.013
[16]   COATS A W, REDFERN J P Kinetic parameters from thermogravimetric data[J]. Nature, 1964, 201 (4914): 68- 69
doi: 10.1038/201068a0
[17]   7D-Soft High Technology Inc. 1st Opt Manual, Release 6.0 (2014)(EB/OL). (2014-01-10) [2018-08-1]. http://www.7dsoft.com/.
[18]   FAN H, ZHANG Y, SU Z, et al A dynamic mathematical model of an ultrasupercritical coal fired once-through boiler-turbine unit[J]. Applied Energy, 2017, 189: 654- 666
[19]   CHEN Z H, ZHU Q J, WANG X, et al Pyrolysis behaviors and kinetic studies on Eucalyptus residues using thermogravimetric analysis[J]. Energy Conversion and Management, 2015, 105: 251- 259
doi: 10.1016/j.enconman.2015.07.077
[20]   SUNGSUK P, CHAYAPOM S, SUNPHORKA S, et al Prediction of pyrolysis kinetic parameters from biomass constituents based on simplex-lattice mixture design[J]. Chinese Journal of Chemical Engineering, 2016, 24 (4): 535- 542
doi: 10.1016/j.cjche.2016.01.004
[21]   HU Q, YANG H P, XU H S, et al Thermal behavior and reaction kinetics analysis of pyrolysis and subsequent in-situ gasification of torrefied biomass pellets[J]. Energy Conversion and Management, 2018, 161: 205- 214
doi: 10.1016/j.enconman.2018.02.003
[22]   MA J, CHENG J C P Identifying the influential features on the regional energy use intensity of residential buildings based on Random Forests[J]. Applied Energy, 2016, 183: 193- 201
doi: 10.1016/j.apenergy.2016.08.096
[23]   DESPANGE F, MASSART D L Neural networks in multivariate calibration[J]. Analyst, 1998, 123 (11): 157- 178
doi: 10.1039/a805562i
[1] Kai ZHU,Yun-he WANG,Xue-wei QIN,Ya-dong HUANG,Qiang WANG,Ke WU. Effect of heating rate on asphalt combustion and gaseous products release characteristics[J]. Journal of ZheJiang University (Engineering Science), 2020, 54(9): 1805-1811.
[2] Dong-xiang KE,Li-min PAN,Sen-lin LUO,Han-qing ZHANG. Android malicious behavior recognition and classification method based on random forest algorithm[J]. Journal of ZheJiang University (Engineering Science), 2019, 53(10): 2013-2023.