Enhanced Joint Estimation of State of Charge and State of Power for Lithium-Ion Batteries Using a Gated Recurrent Unit–Transformer Model With Multiple Input Features

Abstract The estimated results of the state of charge (SOC) for lithium-ion (Li-ion) batteries obtained using conventional data-driven methods often exhibit significant fluctuations that pose challenges in accurately monitoring real-time SOC, which is crucial for ensuring proper discharge operations. Furthermore, the inaccuracies in SOC estimation resulting from these fluctuations also adversely affect the precision of state of power (SOP) estimation. To address the problem of fluctuations in data-driven SOC estimation and enhance SOP accuracy, this work introduces a novel joint estimation strategy leveraging the gated recurrent unit (GRU)–Transformer (G-Trans) model with multiple input features. First, the real-time equivalent open-circuit voltage (EOCV) feature and energy (E) feature are considered for comprehensively representing the internal dynamics of the battery, and thus, they are utilized to construct the multiple input features with voltage, current, and temperature. Second, a G-Trans network is developed by combining GRU and Transformer, where global modeling is performed first, followed by local feature extraction to enhance the accuracy of the model training. In addition, the established multifeatures are used as the input of the G-Trans, which can smooth the fluctuations of the output, leading to improve the SOC estimation accuracy. Finally, to better estimate the SOP over different durations, the peak current under smooth and accurate constraints, voltage constraints, and self-current constraints are integrated to obtain the final peak current, thereby enabling accurate SOP estimation. The experimental results under highway fuel economy test (HWFET) test conditions at 10 °C and 25 °C show that the root mean square errors (RMSE) of SOC estimation are 1.43% and 0.68%, respectively. Meanwhile, for charge/discharge tests with different durations, the SOP estimation accuracy at both 10 °C and 25 °C remains within reasonable ranges while demonstrating stable variation trends.