A Novel Label-Free Supervision Learning Method for Lithium-ion Battery RUL Prediction

Abstract

Lithium-ion battery remaining useful life (RUL) prediction is critical for battery health management. Machine-learning-based method is often used to predict battery RUL, an accurate prediction is dependent on a large amount of labeled data, which is difficult and expensive to obtain. This paper proposes a new method to train the data-model of battery RUL prediction with constraint derived from prior physical knowledge. The constraint specifies the nonlinear function between the battery RUL and energy-throughput. Box-Cox transformation (BCT) is utilized to optimize the constraint, and transform the nonlinear function into a linear one. Then the physical knowledge function between the energy-throughput and RUL is constructed to generate new labeled data, which are used to train Artificial Neural Network (ANN) to achieve the label-free supervision battery RUL prediction data-model. The experimental results demonstrate that proposed method effectively reduces the labeled data under the premise of ensuring the accuracy of the prediction result.