Abstract:Accurate and rapid prediction of the lifetime is essential for accelerating the application of ultracapacitors. To overcome the large inconsistencies in the lifetime of ultracapacitors, an end-to-end remaining useful life (RUL) prediction method based on the convolutional neural network (CNN) is proposed. It directly establishes the mapping between the charging and discharging data collected within a few consecutive cycles and the corresponding remaining useful life. It learns many ageing features from limited raw data without any expert knowledge. While improving the prediction accuracy of the RUL, the required test time drops greatly. Validation results based on 113 ultracapacitors demonstrate that our method can accurately predict RUL by using the data within 5 consecutive cycles collected at any ageing stage, and the root mean square error is 501 cycles. Our method demonstrates higher accuracy compared with conventional feature-based prediction methods, while required input data are sharply reduced. Such 5-cycle testing can be conducted within 15 min to collect enough data for RUL prediction. Our work highlights the promise of data-driven approaches to predict the degradation of energy storage devices.

Rapid Ultracapacitor Life Prediction with a Convolutional Neural Network