A data reconstruction-based Monte Carlo method for remaining useful life prediction of lithium-ion battery with few historical samples

Abstract

Lithium-ion battery (LIB) has been widely used in many energy storage systems, and the accurate remaining useful life (RUL) prediction of LIB is essential to ensure the safe operation of these systems. However, for the LIB installed in a new system, there are only few historical samples available for RUL prediction, which makes it difficult to build accurate RUL prediction model for LIB. Therefore, this paper proposes a data reconstruction-based Monte Carlo method to solve the problems caused by few historical samples. First, a Wiener process-based data reconstruction algorithm is used to reconstruct historical dataset, providing sufficient prior information for the Monte Carlo method. Then, Bayesian theory and a parameter update scheme are proposed to update the reconstructed dataset and the corresponding parameter distributions, so that the training data of RUL prediction model can be closer to the data of target LIB. Finally, long short-term memory neural network is applied for RUL prediction. The effectiveness of our model is verified by two real LIB datasets. Compared with some existing RUL prediction models of LIB, the proposed model has better generality and higher prediction accuracy under the condition of few historical samples.