A thermo-electrical dual control strategy for SOEC system based on a neural network feedforward algorithm

Solid oxide electrolysis cells (SOECs) offer a promising approach to converting renewable energy into syngas through co-electrolysis, enabling efficient energy storage. However, the inherent variability of renewable energy sources, combined with the complex interactions among multiphysics fields and components within the SOEC system, poses a significant challenge to achieving rapid dynamic regulation. This paper develops a comprehensive model of the SOEC system, including the evaporator, electric heater, heat exchanger, and SOEC stack. Through a detailed multi-time-scale characteristic analysis, it is found that the fuel flow rate exhibits advantages in controlling the stack temperature and voltage. Then the effects of basic fuel flow control (FFC), air flow control (AFC) and constant conversion rate control (CCRC) on key performance are compared. The results indicate that while FFC maintains high system efficiency and demonstrates significant advantages in regulating stack temperature, it also induces substantial voltage fluctuations during the adjustment process. To address this issue, a feedforward control strategy based on a Levenberg-Marquardt neural network is proposed, aiming to improve the SOEC's overall transient performance across time scales through precise regulation of the fuel flow rate. Step tests indicate that, compared with conventional FFC, the proposed control algorithm not only further enhances temperature regulation but also reduces voltage fluctuations from 0.95 V to 0.23 V, thereby achieving dual thermo-electrical control for the SOEC system. The effectiveness of this approach is further validated under actual photovoltaic current conditions. • A control-oriented dynamic model of the SOEC system is developed. • The multi-time-scale characteristic of the SOEC system is analyzed in detail. • Compare effects of air flow, fuel flow, and fixed conversion control on the system. • Establish a neural network feedforward-based thermoelectric dual control strategy. • The strategy stabilizes temperature and cuts voltage swings from 0.95 V to 0.23 V.