State of Charge Estimation of Lithium-Ion Batteries Using Convolutional Neural Networks in Electric Vehicle Applications

The emergence of Electric Vehicles (EVs) has introduced new issues and complexities, particularly in accurately gauging the remaining battery capacity or the distance an EV can travel before needing a recharge, known as the State of Charge (SoC) estimation problem. This study aims to develop an innovative solution to this problem while considering the usual computational complexity requirements. It emphasizes maintaining low memory and computational demands for practical implementation within a Battery Management System (BMS). By leveraging contemporary tools from artificial intelligence, especially Machine Learning (ML) models, the findings indicate that accurate SoC prediction can be achieved with simple ML models, given a well-curated dataset. Specifically, a Convolutional Neural Network (CNN) with two hidden layers and a limited number of hidden neurons shows promising results in estimating the SoC for a lithium-ion battery cell while maintaining low computational complexity. At 25°C, the CNN model reports a mean absolute error (MAE) of 0.97 and a root mean squared error (RMSE) of 1.71. This research offers significant potential for future advancements in SoC estimation, providing a solid foundation for addressing the issue and suggesting an optimal ML architecture for future BMSs to improve the SoC prediction accuracy.