Accurate State of Charge (SOC) estimation is crucial for lithium‐ion battery management under complex dynamic conditions and varying temperatures. This study proposes a novel deep learning framework, ResNet‐GRU‐Attention, which integrates the temporal feature extraction of gated recurrent units (GRU), the dynamic focusing of the attention mechanism, and the error correction of residual networks (ResNet). The model is comprehensively evaluated on multiple datasets covering various driving cycles and temperatures, and compared against several baselines. Results demonstrate that the proposed model consistently achieves superior performance across all tested conditions. Under room‐temperature dynamic cycles, its mean absolute error (MAE) and root mean squared error (RMSE) are typically below 1.63% and 2.00%, respectively. Even under the demanding low‐temperature DST cycle, it maintains a clear margin over all baselines. Compared to the baseline GRU model, the RMSE reduction ranges from 16% to 58% across different operating conditions and temperatures, highlighting its advantages in accuracy and robustness.
Deng et al. (Mon,) studied this question.