Research on voltage inconsistency diagnosis of power battery based on PSO-VMD-improved local outlier factor

The performance inconsistency of cells in power battery packs is influenced by multiple factors, including manufacturing processes and operating environments. Voltage, as the most intuitive indicator of inconsistency, serves as a critical parameter for diagnostics. However, existing diagnostic methods fail to effectively decouple high-frequency components from voltage data, which limits diagnostic accuracy. To address this issue, this research proposes a two-stage hybrid diagnostic method based on real-world vehicle data for decoupling and abnormal cell localization. First, the hyperparameters of variational mode decomposition (VMD) are optimized using a collaborative optimization mechanism that combines the particle swarm optimization (PSO) algorithm and Kullback-Leibler (KL) divergence. Second, the local outlier factor (LOF) algorithm is improved by introducing a calculation window and data normalization, while its hyperparameters are optimized through grid search. Finally, an improved VMD algorithm is integrated with the LOF algorithm to establish a novel diagnostic approach for battery inconsistency abnormalities. Experimental results show that the optimized VMD effectively decouples high-frequency components and extracts mid-to-low-frequency components reflecting battery performance, while successfully identifying early thermal runaway signs in cell #81. This approach not only suppresses noise interference but also enhances diagnostic accuracy, providing reliable technical support for the safe operation of power battery systems.