SOC estimation for lithium-ion batteries based on BiGRU with SE attention and Savitzky-Golay filter

The dynamic and diverse working condition leads to the unstable internal performance of lithium batteries, resulting in diminished accuracy in state of charge (SOC) estimation. To address this problem, the bidirectional gated recurrent unit (BiGRU) network with the Squeeze-and-Excitation (SE) attention mechanism and Savitzky-Golay (SG) filtering was proposed to improve the accuracy of SOC estimation. Firstly, BiGRU was utilized to capture bidirectional features, offering a more comprehensive consideration of past and future contextual information. Subsequently, the SE attention mechanism was applied to enhance the weight of useful information, by employing a squeezing operation to capture global spatial information and utilizing an excitation process to further explore nonlinear dependencies between channels. Additionally, the SG filter was employed to fit polynomial functions for smoothing waveforms to enhance the accuracy of SOC estimation. Experiments with the public LiFePO4 battery dataset demonstrated the effectiveness of the developed method. Compared to existing methods, our approach achieved superior prediction accuracy, with MAE below 1 %, MSE below 0.017 %, and RMSE below 1.5 %.