Activation energy prediction of biomass wastes based on different neural network topologies

Cepeliogullar O., Mutlu İ., Yaman S., Haykiri-Acma H.

Fuel, vol.220, pp.535-545, 2018 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 220
  • Publication Date: 2018
  • Doi Number: 10.1016/j.fuel.2018.02.045
  • Journal Name: Fuel
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.535-545
  • Keywords: Biomass, Pyrolysis, TGA, Activation energy, Artificial Neural Network (ANN), CO-PYROLYSIS, KINETIC-PARAMETERS, GASIFICATION, SIMULATION, BIOCHAR, MODEL, TGA
  • Istanbul Technical University Affiliated: Yes


© 2018 Elsevier LtdThe present paper discusses the thermal data prediction performance of ANN for more than one biomass as well as the reliability of this ANN predicted data in the further steps. Lignocellulosic forest residue (LFR) and olive oil residue (OOR) were selected as biomass feedstocks. The thermal data prediction performance of ANN was performed based on two approaches by developing; i) two individual networks for each feedstock, and ii) one-network for both feedstocks. After fixing the main structure of the networks, optimization studies were carried out to determine the best network configuration. In this way, it was also aimed to discuss the effect of internal ANN parameters to the overall prediction capability for more complex problems. At the final step, the predicted data was applied to calculate the activation energies based on three conventional kinetic models and the results were compared with the ones calculated using experimental thermal data. In the end, it was concluded the experimental thermal data fitted quite well to the ANN predicted data (R2 > 0.99) but more complex network topology was required for combined network due to the complexity of the dataset. Most importantly, it is shown that the predicted data can be applicable for the further steps such as in the calculation of the activation energies using different models.