Analysis and prediction of solid oxide electrolysis stack's aging trajectory considering nonstationary phenomenon based on VMD-TCN-LSTM

Degradation prediction of the solid oxide electrolysis cell (SOEC) is critical to extend its lifetime. However, SOEC is a nonlinear time-varying system and suffers from large fluctuations in the output signal due to nonstationary high temperature and humidity, as well as other factors that jointly affect performance, resulting in insufficient prediction accuracy. In this paper, a combined prediction method that integrates variational mode decomposition (VMD), time convolutional network (TCN) with long short-term memory network (LSTM) is proposed. Current signals are first decomposed into relatively stable components using the VMD technique. And then, TCN that can mine temporal feature information is combined with LSTM to predict the time series mode components. Comparative results with other base models show that the proposed method improves the accuracy of SOEC aging trajectory prediction by 85.69 %. This highlights the significance of signal processing technique, which provides an efficient and reliable solution for SOEC performance prediction. • Current signals with different degrees of fluctuation are measured and analyzed. • Noisy dataset has significant impact on SOEC stack's aging trajectory prediction. • VMD-TCN-LSTM is constructed for solving nonstationary aging prediction problem.