Remaining useful life prediction of lithium-ion battery with nonparametric degradation modeling and incomplete data

The accurate and reliable prediction of remaining useful life (RUL) plays a crucial role in ensuring the safe operation of batteries. Most existing RUL prediction methods are based on complete and dense monitoring data. However, in real-world applications, monitoring data often exhibit incompleteness, i.e., sparse data or fragment data, especially in intricate and harsh operating conditions, which brings great challenges to the accurate RUL prediction of batteries. A nonparametric degradation modeling method has been proposed in our previous work to deal with the RUL prediction issue using fragment data. The basic idea is to construct a state-dependent degradation model automatically driven by the degradation data via functional principal component analysis (FPCA). This method transforms the RUL prediction issues into an iterative optimization problem. This paper further develops this nonparametric degradation modeling method and apply it into the RUL prediction of lithium-ion batteries with incomplete data. The major enhancements are as follows. A new Pseudo-Huber loss function is employed in the gradient descent optimization to improve the accuracy of RUL prediction. An iterative optimization algorithm combined with Bayesian updating is proposed to estimate the distribution of functional principal component (FPC) scores, which provides the uncertainty of the RUL instead of only providing the point estimation. The proposed method is evaluated using the Toyota-MIT-Stanford battery experimental data. The results demonstrate that the proposed method provides more accurate and reliable prediction results than other prognostic methods.