State-of-health prediction of lithium-ion power batteries via incomplete tensor representation learning

<p indent="0mm">State of health (SOH) prediction of lithium-ion power batteries is central to battery health management and is essential for ensuring the reliability and safety of battery-operated systems. Data-driven SOH prediction methods have been widely adopted due to their high flexibility and the absence of a need for detailed electrochemical modeling. However, existing data-driven approaches typically rely on complete battery performance data (e.g., current, voltage, and impedance). In real-world operating conditions, unavoidable issues such as sensor malfunctions often lead to incomplete, noisy, or inconsistent performance data, which severely challenge accurate SOH prediction. To address this problem and achieve reliable SOH prediction under incomplete performance data, this paper proposes a lithium-ion power battery SOH prediction method based on incomplete tensor representation learning. First, in the tensor representation learning module, battery performance data are modeled as a fourth-order incomplete tensor with the modes “battery cell × sampling time × monitoring indicator × cycle index”, and an incomplete tensor representation learning model is constructed to obtain complete feature representations. Second, in the feature extraction module, three factor matrices and a cycle-indicator feature tensor are introduced, and informative features are extracted through mode-wise tensor-matrix multiplication. Finally, in the performance prediction module, a Transformer-based model is employed to predict battery performance over the next |<italic>K</italic>| cycles, enabling accurate SOH prediction. Experimental results on two widely used battery performance datasets show that the proposed method can accurately predict battery performance data in future cycles based on incomplete historical monitoring data, thereby achieving accurate SOH prediction.