Performance Optimization and Selection of Operating Parameters for a Solid Oxide Fuel Cell Stack

Abstract

In this study, a model of current densities for a tencell solid oxide fuel cell (SOFC) stack is learned and developed due to the utilization of an improved backpropagation neural network (BPNN). To build the learning data of the BPNN, the operating parameters are suitably arranged by the Taguchi orthogonal array, which totals seven factors with five levels, respectively, that act as the inputs of BPNN. Also, the average current densities for the tencell SOFC stack achieved by the numerical method act as the outputs of the BPNN. The effectiveness of the developed BPNN mathematical algorithm to predict performance of the SOFC stack is proved by the learning errors smaller than 0.11% and the predicting errors less than 0.52%. Then, the calculating algorithms of the BPNN are adopted to proceed with the optimization based on the electrical performance of the sum of the average current densities for the tencell SOFC stack. Thus, the best and the worst performances are found to be Fmax = 57795.622 Am−2 and Fmin = 33939.362 Am−2, respectively. It is also the operating window of the performance for the SOFC stack developed by the improved BPNN. Furthermore, an inverse predicting model of the SOFC stack is developed by the calculating algorithms of the BPNN. This model is proved to effectively predict the operating parameters to achieve a desired performance output of the SOFC stack. Combination of these calculating algorithms developed by the improved BPNN gives the possibility to complete dynamic control of the operating parameters, such as the mole fraction of species and mole flow rate in the inlet, which are considered to be changeable.