A Soh and Rul Prediction Model for Supercapacitors Based on Integrated Model Optimization Algorithm

Abstract

The rapid increase in demand for energy, particularly electrical power, in modern society is remarkable. As an intermediary in energy production and consumption, energy storage devices have seen growing emphasis on their safety and stability, making these aspects crucial. Supercapacitors, known for their high-power density and long cycle life, have been widely applied across various fields. The state of health (SOH) and remaining useful life (RUL) are key indicators for their continued usability. This paper proposes a hybrid prediction model for supercapacitor SOH and RUL based on a Convolutional Neural Network (CNN) and Bidirectional Gated Recurrent Unit (BiGRU). Additionally, a model optimization algorithm leveraging reinforcement learning was employed to enhance performance. Validation results on two datasets indicate that the optimized model achieved significantly improved predictive accuracy, with reductions of 30.68%, 45.64%, 36.76%, and 16.75% in MSE, MAE, MAPE, and RMSE, respectively.

Keywords: Reinforcement learning algorithm, Bayesian optimization algorithm, Supercapacitor, state of health, Remain useful life, Multi-armed Bandit algorithm