A novel hybrid framework for SOC estimation using PatchMixer-LSTM and adaptive UKF

Accurate state-of-charge (SOC) estimation remains a critical challenge in battery management systems (BMS) for electric vehicles (EVs), primarily due to the nonlinear dynamics of lithium-ion batteries and the presence of measurement noise. This study proposes a novel hybrid framework that integrates PatchMixer-LSTM with an adaptive unscented Kalman filter (AUKF) to improve SOC estimation under complex operating conditions. The PatchMixer-LSTM architecture leverages depthwise separable convolutions and LSTM units to capture both local spatial features and long-term temporal dependencies in multivariate battery data. Concurrently, the AUKF dynamically adjusts its parameters to filter measurement noise, thereby improving estimation robustness. Validation under the US06 and FUDS driving cycles across varying temperatures demonstrates the effectiveness of the proposed approach. Specifically, the proposed PatchMixer-LSTM-AUKF framework achieves a root mean square error (RMSE) of 0.12% and a maximum error (MAXE) of 0.40% at 50 °C on the US06 cycle, significantly outperforming EI-LSTM-CO (RMSE: 0.44%, MAXE: 1.24%) and Autoformer-AUKF (RMSE: 1.60%, MAXE: 8.99%). The proposed hybrid framework reduces error accumulation and improves robustness under diverse operating conditions, enabling a more reliable and high-precision implementation.