Towards Real-Time Anomaly Detection of Lithium-Ion Battery Voltage: A Novel Online Framework for Electric Vehicles

Abstract Detecting voltage anomalies in lithium-ion batteries is crucial for ensuring the safe and stable operation of electric vehicles. However, accurately determining the terminal voltage of individual cells during normal operation and reliably identifying voltage abnormalities present significant technical challenges. To address the limitations of model-based approaches that rely on prior experimental data or neural network methods with high computational overhead, this paper proposes an analytical framework that incorporates concise bias parameters to estimate the terminal voltage of each cell with high accuracy yet low complexity. Building upon this estimator, an error-sum exponent is defined to quantify the discrepancy between measured and estimated terminal voltages; and the polarity of this exponent enables real-time online discrimination of cell-level voltage anomalies. A mathematical optimization method for the error-sum window length is further derived to facilitate its real-time adaptation. Experimental validation using real-world vehicle data demonstrates that the proposed method achieves both a mean absolute error and a root-mean-square error in terminal-voltage estimation of no more than 0.015 V. In online deployment, it accurately detects voltage anomalies in real-vehicle data with substantial noise, providing an efficient and lightweight solution for the real-time safety monitoring of large-scale traction battery systems.