State of Charge Prediction of Lithium-Ion Batteries Based on Artificial Neural Networks and Reduced Data

Lithium-ion batteries (LIBs) are the key technology for the electrification of the transport sector. Since LIBs have a complex, electrochemical structure, it is a challenge to accurately determine the condition, which is crucial for safety and efficiency during operation. This paper presents the forecasting of the state-of-charge (SOC) of a LIB based on machine learning (ML) algorithms. Data from battery simulation and augmented data are additionally used to train the models. To reduce the dimension of the feature matrix, a singular value decomposition is performed. A multi-layer perceptron (MLP) and a convolutional neural network (CNN) are compared to a linear regression. The impact of the augmented data on the prediction accuracies and the reliabilities of the models is analyzed. The lowest test error is achieved using the CNN with augmented data with a root mean square error (RMSE) of 1.78 %. The results show the applicability of data-driven models for the SOC prediction and the optimization potential using data augmentation techniques.