State-of-Charge Estimation of Lithium-Ion Batteries Using Machine Learning Based on Augmented Data

Abstract

For enhanced safety and efficiency, it is crucial to accurately monitor and determine the condition of a lithium-ion battery (LIB). Because of dynamic operating conditions and several diverse aging mechanisms, it is necessary to further improve the algorithms of the battery management system. In this paper, machine learning (ML) is used to estimate the state of charge (SOC). As data-driven models highly depend on their data basis and battery tests are time and cost expensive, the data is enriched by simulation and augmented data. Based on this data, a linear regression, a support vector machine, and different types of artificial neural networks are trained. All models are tested on a real world data set. A lowest test error is obtained using the convolutional neural network with a test RMSE of 1.74 %. The results highlight the accuracy of data-driven models for the prediction of the battery state and the potential to improve the models by using data augmentation techniques.