The issues of unstable capacity regeneration, discharge capacity fluctuations, and significant individual variability during the charge–discharge process of supercapacitors pose challenges to accurately predicting their state of health (SOH), reducing their efficiency and safety. To address these challenges, this paper proposes a supercapacitor state prediction model based on complete ensemble empirical mode decomposition with adaptive noise and bidirectional gated recurrent units (BiGRU). This approach combines the advantages of signal decomposition in feature extraction with the ability of BiGRU to process long‐term time‐series data. To maximize the model's predictive performance, the optimization problem is reformulated as a sequential decision‐making task. A fusion model optimization algorithm, integrating reinforcement learning with model optimization techniques, is proposed to optimize key hyperparameters of the model. An aging test platform for supercapacitors is developed, and the proposed method is validated using rigorously partitioned experimental and public datasets. By isolating high‐frequency fluctuations via signal decomposition, the model effectively mitigates the impact of capacity instability. The results demonstrate that the optimized model achieves significantly higher prediction accuracy, better generalization, and improved robustness compared to the unoptimized model and traditional prediction models. The proposed method not only enhances the economic efficiency and safety of supercapacitor applications but also holds significant potential for expanding practical use in various scenarios.
Wang et al. (Wed,) studied this question.