Multidimensional Consistency Diagnosis of Power Batteries Based on Optimized Modal Decomposition and Unsupervised Learning

To address the efficiency degradation and safety risks caused by power battery inconsistency, as well as limitations of existing diagnostic methods in real vehicle applications such as strong noise sensitivity, heavy parameter dependence, and single-feature, this study proposes a multidimensional unsupervised diagnostic strategy that integrates gray wolf optimization-based variational mode decomposition (GWO-VMD) with Isolation Forest (IF). The proposed method applies the GWO algorithm to adaptively optimize key VMD parameters and extracts low-frequency voltage components that reflect intrinsic battery degradation trends from noisy real-world vehicle data. On this basis, a multidimensional feature space is constructed by integrating the robust coefficient of variation (CV), Jensen–Shannon (JS) divergence, minimum voltage, and voltage range. The robust CV eliminates dynamic load disturbances and characterizes time-domain stability, while the JS divergence quantifies differences in the probability distribution shape. Based on the orthogonal decoupling characteristics of stable aging in the feature space and the multidimensional coupled evolution behavior exhibited by substantive faults, the IF algorithm achieves isolation and quantitative scoring of abnormal cells. Validation using data from two real electric vehicles (EVs) demonstrates that the proposed model effectively suppresses dynamic operating condition interference and accurately identifies evolutionary faulty cells with cumulative electrochemical heterogeneity. Ablation experiments confirm that GWO-VMD processing significantly improves the fault detection sensitivity and robustness. Finally, a collaborative framework of cloud-based parameter optimization and vehicle-side real-time diagnosis is proposed, which provides effective technical support for consistency management and preventive maintenance throughout the full life cycle of power batteries.