A comprehensive framework of synchronous SOC-SOH joint estimation for lithium-ion battery with multi-depth expert networks

Accurate joint estimation of state of health (SOH) and state of charge (SOC) plays a vital role in improving the reliability of intelligent battery management systems. However, existing approaches often fail to ensure the synchronized estimation of SOC and SOH. In addition, reliable SOH estimation typically relies on complex features available only during specific charging intervals, reducing its practicality. Motivated by these challenges, this study proposes a comprehensive framework for joint and synchronized estimation of lithium-ion battery SOC and SOH with multi-depth expert networks. To accommodate the differing feature scale requirements of SOC and SOH, task-specific and shared feature extraction networks with varying depths are designed. During the fusion stage, a Mamba module is introduced to capture global patterns from the coarse-grained time series relevant to SOH, while an LSTM is employed to extract fine-grained local features critical to SOC. Finally, a dynamically weighted joint loss, guided by intermediate feature interactions, is proposed to enhance estimation accuracy. The validation results on the MIT degradation dataset demonstrate that the proposed method reliably achieves synchronized SOC-SOH joint estimation across various operating conditions throughout the battery's entire lifecycle, exhibiting strong generalization capabilities. The root mean squared errors for both SOC and SOH estimates remain below 1.7 %, representing reductions of at least 57.4 % and 18.6 % respectively compared to other methods.