A hybrid SVR-PSO model to predict a CFD-based optimised bubbling fluidised bed pyrolysis reactor

Salman Jalalifar, Mojtaba Masoudi, Rouzbeh Abbassi*, Vikram Garaniya, Mohammadmahdi Ghiji, Fatemeh Salehi

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

39 Citations (Scopus)


Comprehensive scrutiny is necessary to achieve an optimised set of operating conditions for a pyrolysis reactor to attain the maximum amount of the desired product. To reach this goal, a computational fluid dynamic (CFD) model is developed for biomass fast pyrolysis process and then validated using the experiment of a standard lab-scale bubbling fluidised bed reactor. This is followed by a detailed CFD parametric study. Key influencing parameters investigated are operating temperature, biomass flow rate, biomass and sand particle sizes, carrier gas velocity, biomass injector location, and pre-treatment temperature. Machine learning algorithms (MLAs) are then employed to predict the optimised conditions that lead to the maximum bio-oil yield. For this purpose, support vector regression with particle swarm optimisation algorithm (SVR-PSO) is developed and applied to the CFD datasets to predict the optimum values of parameters. The maximum bio-oil yield is then computed using the optimum values of the parameters. The CFD simulation is also performed using the optimum parameters obtained by the SVR-PSO. The CFD results and the values predicted by the MLA for the product yields are finally compared where a good agreement is achieved.

Original languageEnglish
Article number116414
Pages (from-to)1-12
Number of pages12
Publication statusPublished - 15 Jan 2020


  • Support vector regression (SVR)
  • Particle swarm optimisation (PSO)
  • Computational fluid dynamic (CFD) simulation
  • Bubblingfluidised bed reactor
  • Fast pyrolysis process


Dive into the research topics of 'A hybrid SVR-PSO model to predict a CFD-based optimised bubbling fluidised bed pyrolysis reactor'. Together they form a unique fingerprint.

Cite this