Advanced search
2 files | 14.00 MB Add to list

Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data

Author
Organization
Abstract
Computational fluid dynamics (CFD) is an effective tool to investigate biomass fast pyrolysis in fluidized bed reactor for bio-oil production, while it requires huge computational time when optimizing operating conditions or simulating large/industrial units. Machine learning (ML) is a promising approach to achieving both accuracy and efficiency. In this work, a reduced-order model including long short-term memory (LSTM) layer, pooling layer, and fully connected layer was established to predict future mass flow rates by training the historical CFD data. Unsteady mass flow rates, which are normally used to determine product yields, were treated as data of time series in ML. Influencing factors, such as sequence length, number of neurons, learning rate, subsequences order (shuffle or not), number of LSTM layers, and ratio of testing set, were evaluated to obtain their optimal values. The developed LSTM model framework and training process showed good applicability for the dataset of different species and temperatures. Product yields predicted by the derived LSTM were in good agreement with those obtained by CFD, while nearly 30% computational effort was saved. Thus, it is clearly seen that the well-predicted fluctuating characteristics and final product yields are helpful to improve accuracy of process simulation for digitalizing key reactors and building smart factories.
Keywords
Bio-oil, Biomass fast pyrolysis, Fluidized bed, CFD, Machine learning, LSTM

Downloads

  • (...).pdf
    • full text (Published version)
    • |
    • UGent only
    • |
    • PDF
    • |
    • 11.06 MB
  • manucript.pdf
    • full text (Accepted manuscript)
    • |
    • open access
    • |
    • PDF
    • |
    • 2.93 MB

Citation

Please use this url to cite or link to this publication:

MLA
Zhong, Hanbin, et al. “Prediction of Instantaneous Yield of Bio-Oil in Fluidized Biomass Pyrolysis Using Long Short-Term Memory Network Based on Computational Fluid Dynamics Data.” JOURNAL OF CLEANER PRODUCTION, vol. 391, 2023, doi:10.1016/j.jclepro.2023.136192.
APA
Zhong, H., Wei, Z., Man, Y., Pan, S., Zhang, J., Niu, B., … Xiong, Q. (2023). Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data. JOURNAL OF CLEANER PRODUCTION, 391. https://doi.org/10.1016/j.jclepro.2023.136192
Chicago author-date
Zhong, Hanbin, Zhenyu Wei, Yi Man, Shaowei Pan, Juntao Zhang, Ben Niu, Xi Yu, Yi Ouyang, and Qingang Xiong. 2023. “Prediction of Instantaneous Yield of Bio-Oil in Fluidized Biomass Pyrolysis Using Long Short-Term Memory Network Based on Computational Fluid Dynamics Data.” JOURNAL OF CLEANER PRODUCTION 391. https://doi.org/10.1016/j.jclepro.2023.136192.
Chicago author-date (all authors)
Zhong, Hanbin, Zhenyu Wei, Yi Man, Shaowei Pan, Juntao Zhang, Ben Niu, Xi Yu, Yi Ouyang, and Qingang Xiong. 2023. “Prediction of Instantaneous Yield of Bio-Oil in Fluidized Biomass Pyrolysis Using Long Short-Term Memory Network Based on Computational Fluid Dynamics Data.” JOURNAL OF CLEANER PRODUCTION 391. doi:10.1016/j.jclepro.2023.136192.
Vancouver
1.
Zhong H, Wei Z, Man Y, Pan S, Zhang J, Niu B, et al. Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data. JOURNAL OF CLEANER PRODUCTION. 2023;391.
IEEE
[1]
H. Zhong et al., “Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data,” JOURNAL OF CLEANER PRODUCTION, vol. 391, 2023.
@article{01GTSBE048D73RQK50T43ZXSYG,
  abstract     = {{Computational fluid dynamics (CFD) is an effective tool to investigate biomass fast pyrolysis in fluidized bed reactor for bio-oil production, while it requires huge computational time when optimizing operating conditions or simulating large/industrial units. Machine learning (ML) is a promising approach to achieving both accuracy and efficiency. In this work, a reduced-order model including long short-term memory (LSTM) layer, pooling layer, and fully connected layer was established to predict future mass flow rates by training the historical CFD data. Unsteady mass flow rates, which are normally used to determine product yields, were treated as data of time series in ML. Influencing factors, such as sequence length, number of neurons, learning rate, subsequences order (shuffle or not), number of LSTM layers, and ratio of testing set, were evaluated to obtain their optimal values. The developed LSTM model framework and training process showed good applicability for the dataset of different species and temperatures. Product yields predicted by the derived LSTM were in good agreement with those obtained by CFD, while nearly 30% computational effort was saved. Thus, it is clearly seen that the well-predicted fluctuating characteristics and final product yields are helpful to improve accuracy of process simulation for digitalizing key reactors and building smart factories.}},
  articleno    = {{136192}},
  author       = {{Zhong, Hanbin and Wei, Zhenyu and Man, Yi and Pan, Shaowei and Zhang, Juntao and Niu, Ben and Yu, Xi and Ouyang, Yi and Xiong, Qingang}},
  issn         = {{0959-6526}},
  journal      = {{JOURNAL OF CLEANER PRODUCTION}},
  keywords     = {{Bio-oil,Biomass fast pyrolysis,Fluidized bed,CFD,Machine learning,LSTM}},
  language     = {{eng}},
  pages        = {{12}},
  title        = {{Prediction of instantaneous yield of bio-oil in fluidized biomass pyrolysis using long short-term memory network based on computational fluid dynamics data}},
  url          = {{http://doi.org/10.1016/j.jclepro.2023.136192}},
  volume       = {{391}},
  year         = {{2023}},
}

Altmetric
View in Altmetric
Web of Science
Times cited: