Abstract: [Objectives] In order to solve the problems of of capacity regeneration，unstable discharge capacity，and significant individual variability during the charging and discharging processes of supercapacitors，a state of health (SOH) prediction model based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and bidirectional gated recurrent unit (BiGRU) neural networks was proposed to enhance the safety and efficiency of supercapacitor applications.Subsequently，a fusion optimization algorithm was designed by integrating the Q-learning algorithm with Bayesian optimization and applied to the constructed model.[Methods] The CEEMDAN method was used to process the discharge capacity sequence，with the resulting residual components serving as the model's output.The input features of the model include the charging and discharging time and the voltage sequence during the discharge phase.The model's performance was evaluated using six metrics: mean absolute error，mean squared error，root mean squared error，mean absolute percentage error，maximum error，and coefficient of determination.[Results] Compared to the unoptimized model，the optimized model demonstrated significantly reduced errors across five evaluation metrics，while he coefficient of determination shows a slight improvement.[Conclusions] The proposed prediction model exhibits strong robustness and high prediction accuracy，making it highly significant for expanding its potential application scenarios.

Key words: energy storage, supercapacitor, empirical mode decomposition, bidirectional gated recurrent unit, Bayesian optimization, state of health, optimization algorithm, prediction model