Performance Prediction and Evaluation of Biogas-Fed Solid Oxide Fuel Cell Plants Based on the Spatial and Temporal...

The solid oxide fuel cell (SOFC) system as green and efficient power-generation equipment needs to match a reasonable system design and parameter sensitivity analysis in different application fields. The complex SOFC system is accompanied by electrochemical, ionic conduction, and mass transport reactions, making it difficult to predict stack performance. In this article, the conceptual design of the electrolyte-supported SOFC with a biogas-fed cogeneration system is carried out. Two system layouts featuring different anode recirculation schemes (hot and cold) and reforming methods under fluctuation fuel types are designed. Based on the above simulation data, three different algorithm models, including long short-term memory (LSTM), one-dimensional convolutional neural network (1DCNN), and integration model, are used to estimate the SOFC response and predict the SOFC voltage under fluctuating fuel inlet conditions with multiple evaluation criteria. The CatBoost integrator effectively combines the characteristics of LSTM and 1DCNN in the spatial and temporal information, breaks the barriers between temporal and nontemporal data, and avoids the suboptimal solution of static fusion, further improving the prediction accuracy and reliability for the cold recirculation plant configuration scheme. The determination coefficient R ² improves from 0.5 to 0.8, the RMSE and MAE index decrease by up to 44%, and the MAPE index decreases by up to 29%. For the hot recirculation scheme, the fusion algorithm also shows better prediction results under strong periodic training data. Thus, the proposed prediction model can effectively evaluate the SOFC system output without conducting additional physical experiments.