A dual-stage pseudo-label guided state of charge estimation method for lithium-ion batteries

Accurate lithium-ion (Li-ion) battery State of Charge (SOC) estimation is crucial for ensuring their safe and reliable operation. However, the SOC of the battery cannot be directly measured using physical techniques and needs to be estimated using measurable data. Moreover, the nonlinear attributes of batteries pose challenges to achieving SOC estimation. To address this issue, a dual-stage pseudo-label guided hybrid model (DSPL) is proposed in this paper to provide accurate SOC estimation results. Firstly, the first stage consists of a Temporal Convolutional Network (TCN), which obtains SOC sequence pseudo labels containing historical information. Then, the pseudo labels are concatenated with the original input to guide the feature extraction in the second stage. Further, a Feature Calibration Module (FCM) is proposed to mitigate noise interference introduced by pseudo labels and emphasize the varying contributions of different features to the SOC estimation. The FCM is combined with position encoding to form a feature-position encoding module that encodes the concatenated input in the entire feature space. Finally, the Transformer model is introduced in the second stage to extract features from the encoded input and estimate the SOC at the current moment. The network is evaluated on the CALCE database, and the MAE obtained by DSPL on the FUDS operating condition at 25 C can reach 0.6650%, proving its outstanding performance.