Hybrid neural network and dynamic decay model for life prediction of solid oxide fuel cell combined heat and power systems

Abstract

The high cost, limited lifespan, poor durability, and low reliability of solid oxide fuel cell (SOFC) combined heat and power systems significantly hinder their widespread adoption in large-scale commercial applications. Additionally, the complex interactions between components within the SOFC system make fault prediction particularly challenging. To address this issue, this study conducts continuous operational tests on two sets of kilowatt-level SOFC systems, collecting performance data. A dynamic response model of the SOFC system is developed on the Simulink platform, systematically analyzing the voltage performance and its dynamic attenuation characteristics. Subsequently, a Kalman filter algorithm is employed to calculate the stack performance attenuation factor (r), which is integrated into the system dynamic model to enable accurate prediction of system-level attenuation. Finally, a neural network model is constructed to effectively capture the performance degradation characteristics of the SOFC system, with a maximum prediction error of 5 %. This hybrid approach, combining the dynamic decay model and the neural network, is used to predict the service life of the SOFC system, with an estimated lifespan of 7750 h for a 40-cell SOFC system. The findings provide an important theoretical foundation and technical support for the optimal design and long-term operation of SOFC systems.