Hybrid Neural Network Model of a Solid Oxide Fuel Cell

  • Miss Bhawasut Chutichai, Department of Chemical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok 10330, Thailand, Thailand
  • Mr Kattiyapon Chaichana, Department of Chemical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok 10330, Thailand, Thailand
  • Miss Dang Saebea, Department of Chemical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok 10330, Thailand, Thailand
  • Dr Suttichai Assabumrungrat, Department of Chemical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok 10330, Thailand, Thailand
  • Dr Moses Tade, Department of Chemical Engineering, Curtin University of Technology, Western Australia 6845, Australia, Australia
  • Dr Amornchai Arpornwichanop, Department of Chemical Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok 10330, Thailand, Thailand

    • A solid oxide fuel cell (SOFC) is considered as a promising energy-conversion device for power generation. The flexibility of using various fuels (i.e., methane, methanol, ethanol, etc.) and the prospect for combined heat and power system are main benefits of SOFC as a result of its high-temperature operation. Modeling of SOFC is an important tool for analyzing and designing fuel cells. In general, the complete description of fuel cells under steady state and dynamic conditions requires an electrochemical model to predict cell electrical characteristics, i.e., the relation of cell voltage and current. However, obtaining the electrochemical model is quite difficult and complicated as it depends on various cell parameters, i.e., operational and structural parameters. The objective of this study is to present a hybrid neural network model of SOFC with direct internal reforming of methane. In the hybrid model strategy, a neural network (NN) was developed to predict a current density based on gas composition and operating temperature and cell voltage. The simulated data obtained from simulations of a detailed electrochemical model was used to train the NN. Then, the developed NN model was integrated with a governing fuel cell model to predict the SOFC performance. The results show that the NN provides a good estimation of the electrical cell characteristics and thus, the hybrid NN and fuel cell model can predict the performance of SOFC. The effect of operating parameters on the SOFC performance in terms of fuel utilization and power density was analyzed using the hybrid NN model.