Hydrogen Safety in Solid Oxide Fuel Cells: an LSTM-Based Model for Predicting Temperature Anomalies and Change Points

Temperature anomalies in solid oxide fuel cells (SOFC) can significantly affect performance, induce thermal stress and evolve towards accident scenario. Accurately predicting these anomalies is critical for maintaining system integrity and safety, possibly providing weak early signals before an incident occurs. This paper presents the development of a predictive model utilizing Long Short-Term Memory (LSTM) networks to forecast temperature anomalies and detect change points between normal and abnormal states in SOFCs. The LSTM model is trained on extensive historical temperature data, capturing temporal dependencies and patterns indicative of potential anomalies. Change point detection mechanisms are integrated to identify transitions between normal and abnormal operating states, enabling timely interventions. The model efficacy in predicting temperature-related issues and detecting change points with high accuracy is verified by extensive runs in a laboratory scale plant. The results indicate that the LSTM-based model significantly outperforms traditional methods in both prediction accuracy and early anomaly detection. The research findings underscore the potential of advanced neural network architectures in predictive maintenance applications, providing a robust tool for managing performances and ensuring operational safety, in hydrogen fuel cell-power generation.