Strong robustness and high accuracy in predicting remaining useful life of supercapacitors

Abstract

Remaining useful life shows extraordinary function in guiding the timely replacement of supercapacitors that reach the service life limit, which has great significance to the security and stability of the energy storage system. In order to more accurately predict the remaining useful life of supercapacitors so as to ensure the reliability of the whole supercapacitor bank, a temporal convolutional network is used. Among them, a residual block can solve the problems of gradient explosion and gradient disappearance, which are widespread in the recurrent neural network. Early stopping technology is used to avoid overfitting, and the Adam algorithm was used to optimize the process of parameter adjustment of the temporal convolutional network. The stability and accuracy of the model prediction were verified by using the capacity attenuation dataset of supercapacitors under different experimental conditions. Meanwhile, to verify the generalization ability of the model, the datasets of supercapacitors at different working conditions without training are input into the temporal convolutional network model. Simulation shows that the temporal convolutional network model exhibits strong robustness and high accuracy in predicting the remaining useful life of supercapacitors.