Power battery voltage inconsistency fault identification method based on DBSCAN and dynamic K-value K-means++ joint clustering algorithm

Abstract As a critical fault in pure electric vehicle (EV) power battery systems, cell inconsistency not only degrades overall battery performance but also poses safety risks. Leveraging vehicle manufacturer big data, this study proposes a hybrid clustering algorithm combining DBSCAN (Density-Based Spatial Clustering of Applications with Noise) and dynamic K-value K-means++ to identify voltage inconsistency faults. The two-stage framework first employs DBSCAN for noise separation and core point extraction, followed by anomaly analysis using dynamic K-means++. Model parameters are optimized using data from 50 alarm-free vehicles with >10,000 km mileage. Fault detection thresholds (Z = 2.58) are established via the Z-score normalization of OF (Outlier Factor) values calculated by the hybrid algorithm. Validation on four vehicles with known inconsistencies demonstrated 12–23 days earlier anomaly detection than existing monitoring systems. Comparative tests confirm superior timeliness over standalone dynamic K-means++ and OPTICS methods, highlighting the algorithm’s engineering applicability.