Multi-Modal Identification in Solid Oxide Fuel Cells Using a Novel Transfer Learning Method

The Solid Oxide Fuel Cell (SOFC) system is a high-power, environment-friendly fixed power generation device with promising future, but the identification of abnormal conditions across different power systems is still challenging. We propose a novel transfer learning (TL) method used for multi-modal identification, which is based on an optimized CNN. Besides, we introduce temporal activation regularization to enhance the learning ability and performance of the model, while using TL to leverage other SOFC system modal mapping knowledge to assist in target model training. Experimental results demonstrate that when data is scarce, the proposed model exhibits great performance with the accuracy of 98.30%.