The LOF-based early warning method achieved a 34-hour lead time, with 95% precision and 93% recall, reducing false alarms by 42% compared to traditional approaches.
A novel LOF-based framework with dynamic thresholding significantly improves detection accuracy and reduces false alarms for EV battery inconsistency.
Absolute Event Rate: 0% vs 0%
An early warning method for detecting inconsistency in electric vehicle (EV) power batteries is proposed. This work introduces a novel LOF-based framework that integrates dynamic thresholding and multi-feature extraction, validated using real-world datasets to improve detection accuracy and reduce false alarms. Battery operation data are segmented through a sliding window mechanism, enabling the extraction of both long-term and short-term voltage anomaly characteristics. Four key indicators-Discrete Fréchet Distance (DFD), Clearance Factor (CLF), Shape Factor (SF), and Shannon Entropy (H) are utilized to characterize uncertainty-related inconsistency features. The LOF algorithm analyzes the constructed feature matrix for anomaly identification, while a dynamic threshold adaptively adjusts the warning criteria under varying operating conditions. The proposed method achieves an average early warning lead time of 34 hours, with precision of 0.95, recall of 0.93, and a false alarm rate 42% lower than conventional LOF-based approaches. Validation across ten EV and e-bike datasets under diverse driving and environmental conditions confirms the method’s strong generalization ability. Experimental results verify that the proposed approach significantly improves detection accuracy, reduces false alarms, and provides timely early warning, offering a reliable and practical solution for safe and consistent EV battery operation.
Xuan et al. (Fri,) reported a other. The LOF-based early warning method achieved a 34-hour lead time, with 95% precision and 93% recall, reducing false alarms by 42% compared to traditional approaches.