Modeling the SOFC by BP neural network algorithm

Abstract

Solid oxide fuel cells (SOFCs) are complex systems in which electrochemistry, thermophysics and ion conduction occur simultaneously. The coupling of the multi-physics field makes the performance of the SOFC hard to predict. Meanwhile, the high operation temperature prevents a fully aware test of the internal conditions of the SOFC. Artificial intelligence (AI) technology requires a small amount of experimental data to build an accurate model of SOFC. In this study, 30 pieces SOFC stack is fabricated and experimentally tested in different furnace temperatures and the BP neural network, support vector machine (SVM)and random forest (RF) are both used to predict the stack performance. Multiple evaluation criteria such as R2, RMSE, MAE, training and testing time are used. The results show that fitting errors of the three algorithms are all within 5%. By analyzing the evaluation criteria, the prediction accuracy, generalization ability and testing time of BP neural network are optimal. The untested data is predicted in order to verify the interpolation of the BP neural network.