An additive attention-enhanced BiGRU model optimized by beluga whale algorithm for SOEC degradation predicting

Performance degradation prediction serves as a critical approach to improving the durability of the solid oxide electrolysis cell (SOEC). Although data-driven methods demonstrate significant potential in performance prediction, their application within SOEC remains limited. In response to this problem, this paper proposes a hybrid deep learning model that integrates bidirectional gated recurring units (BiGRU), an additive attention mechanism, and the beluga whale optimization (BWO) algorithm to improve the accuracy and robustness of SOEC degradation prediction. The framework employs downsampling and Savitzky–Golay filtering for data preprocessing, followed by BiGRU modeling to capture nonlinear degradation patterns, with additive attention enhancing feature focus. The BWO algorithm is used to optimize hyperparameters automatically. Experimental validation demonstrates that the proposed model achieves a prediction accuracy of 99.38 %, with RMSE of 0.0182 and MAE of 0.0136, outperforming several baselines. In remaining useful life prediction, it achieves a remarkably low average error of only 1.29 %, maintaining high accuracy across both early and late degradation stages.