SOC Estimation of a Li-ion Battery using Deep Learning Method: A comparative Study of LSTM and GRU Architecture

The accurate estimation of State of Charge (SOC) is of utmost importance in battery management systems (BMS) due to the complex internal dynamics of batteries, involving self-discharge, capacity loss, and aging. Conventional methods such as Coulomb Counting (CC) and extended kalman filter (EKF) have been employed for SOC estimation; however, they suffer from slow processing speeds and high computational demands. To address these challenges, this study proposes a data-driven approach using state-of-the-art deep learning techniques. Specifically, we investigate the comparative performance of two popular recurrent neural network (RNN) architectures, namely gated recurrent unit (GRU) and long short term memory (LSTM) for SOC estimation. The analysis is based on a comprehensive battery dataset collected across varying temperatures and drive cycles. Encouragingly, both LSTM and GRU models exhibit excellent performance, with the LSTM achieving an root mean squared error (RMSE) of 0.0515 and the GRU achieving an RMSE of 0.02252 at $0^{\circ}\mathrm{C}$. Moreover, the application of deep learning techniques in conjunction with Internet of Things (IoT) devices holds promising potential for enhancing SOC estimation in electric vehicles (EVs). By integrating these advanced models into the EV ecosystem, real-time monitoring and accurate SOC predictions can be achieved, enabling efficient and optimized battery utilization.