Artificial Intelligence Approach in Gasification Integrated Solid Oxide Fuel Cell Cycle


Sezer S., Kartal F., Özveren U.

Fuel, vol.311, 2022 (Journal Indexed in SCI) identifier identifier

  • Publication Type: Article / Article
  • Volume: 311
  • Publication Date: 2022
  • Doi Number: 10.1016/j.fuel.2021.122591
  • Title of Journal : Fuel
  • Keywords: Biomass gasification, Artificial neural network, Solid oxide fuel cell, Bubbling fluidized bed gasifier, Aspen Plus, RESPONSE-SURFACE METHODOLOGY, GIBBS FREE-ENERGY, FLUIDIZED-BED, BIOMASS GASIFICATION, HYDROGEN-PRODUCTION, SYNGAS PRODUCTION, PERFORMANCE EVALUATION, BIODIESEL PRODUCTION, WASTE GASIFICATION, SIMULATION-MODEL

Abstract

With the growing world population and industrial developments, the supply of energy from an economically feasible and widely available source is important. Biomass gasification is a promising technology that produces lower emissions and allows efficient conversion. The gas obtained from the gasification process, especially in steam gasification, consists of a considerable amount of H2 and is used in fuel cells, especially solid oxide fuel cells (SOFC), to generate electricity. SOFC can convert the chemical energy into electricity and is considered as the most suitable fuel cell type for biomass gasification derived fuels. There are numerous research studies on integrated gasification-SOFC systems in the literature. However, these systems are still under development and studies are being conducted on the appropriate design parameters and operating conditions to achieve high energy efficiency. Modeling of the integrated gasification and SOFC system using the thermodynamic method is the simplest way to determine the process behavior. Nowadays, artificial neural networks (ANN) are one of the most popular modeling methods to represent the thermodynamic based gasification and SOFC systems. In this study, an integrated bubbling fluidized bed gasifier and SOFC model was created to generate data for training the ANN models with Aspen Plus simulation. The ANN models predicted the performance parameters in terms of electrical efficiency, net voltage and current density successfully using the varying operating conditions and 30 different biomass types as input parameters. The results showed that the developed ANN models estimated the output parameters with high accuracy by means of R2 greater than 0.999, MAPE < 0.053 and RMSE < 0.751 for training test and validation data sets.