A linear degradation model-based remaining useful life prediction framework of supercapacitors considering non-consistency of degradation trends

Abstract A supercapacitor is a new type of electrochemical energy storage device. Accurately predicting the remaining useful life (RUL) of a supercapacitor is crucial and important for the energy storage management system. This article presents a framework for predicting RUL, integrating a linear degradation model with an adaptive forgetting factor recursive least squares algorithm. Given the degradation trends' nonlinear intrinsic characteristics, the Box-Cox transformation (BCT) is applied to transform the degradation data. Enhancements to the BCT are made using a stepwise linear regression algorithm to enhance its effectiveness. Subsequently, the degradation path of the supercapacitor is characterized using this linear model. From this, a spatial state model of the system is formulated, deriving its basis from the linear degradation approach. The parameters of the linear degradation model are considered as state variables. Iterative updating of the state variables is performed using an adaptive forgetting factor algorithm to minimize the least squares. The algorithm dynamically adjusts the forgetting factor through an incremental PID control algorithm. The incremental PID control algorithm can response to the identification errors of capacitance. After iterations, the linear degradation model is extrapolated to predict RUL. The results indicate that the linear degradation model can explain more than 98% of the capacity fluctuations. The average deviation of RUL predictions for the test set was initially about 5%, decreasing to 0.57% towards the end. Furthermore, the proposed framework offers the advantage of low additional computational costs compared to a nonlinear approach, while still allowing for competitive prediction accuracy.