Interpretable Hybrid CDTCN-GRU Model for Supercapacitor Life Prediction

Supercapacitors are compelling choices for energy storage devices in diverse applications due to their high power density, long lifespan, and wide operating temperature range. The accurate prediction of the life of supercapacitors plays a vital role in ensuring the reliable operation of these energy storage systems. This study presents a novel hybrid model that combines a custom dilated temporal convolutional network (CDTCN) and a gated recurrent unit (GRU) to predict the remaining useful life (RUL) of supercapacitors. The developed testbed collects real-time experimental data using hall sensors under specific operating conditions from the supercapacitor. The novel CDTCN architecture, with its customized dilations, captures short-term, mid-term, and long-term dependencies via separate branches, while the GRU model captures the sequential and temporal information in the capacity degradation pattern of the supercapacitor module. The proposed CDTCN-GRU model efficiently captures capacity degradation patterns at different current rates for the charging and discharging of the supercapacitor module. The experimental results show that the proposed model demonstrates better performance compared to existing models in RUL estimation, with reduced mean absolute error (MAE) and root-mean-square error (RMSE) values, along with parameter compatibility. The use of explainable artificial intelligence (AI) in the proposed model identifies critical input features for life estimation in supercapacitors and increases the interpretability of the model.