A multi-indicator consistency fusion evaluation method and dual-layer diagnostic strategy based on real vehicle operation data

To address cell consistency degradation and thermal runaway risks in power batteries of electric vehicles (EVs) under complex operating conditions, as well as the absence of cell-level state-of-charge (SOC) information in real-world offline data, this study proposes a cell-level SOC reconstruction method driven by long-term vehicle operational data and a multi-indicator-fusion-based dual-layer graded diagnostic strategy. First, a quasi-open-circuit-voltage (QOCV) reconstruction method is developed to estimate cell-level SOC from long-term operational data. Under normal operating conditions, the reconstructed QOCV–SOC relationships yielded R2 values of 0.98593 and 0.98555 for Cell 1 and Cell 53, respectively. When thermal-runaway data were included, the R2 value for Cell 53 decreased significantly to 0.23804, indicating severe distortion of the voltage–SOC relationship. The reconstructed SOC is then employed to quantify consistency deviations among cells. Subsequently, an improved Local Outlier Factor (ImLOF) algorithm is introduced to detect voltage anomalies while considering local neighborhood characteristics. To comprehensively evaluate battery risk, an AHP-based fusion model combines SOC deviation, voltage ImLOF values, and occurrence frequency into a unified risk score. Finally, a temperature-based safety-gating mechanism is incorporated to distinguish gradual degradation from thermal-runaway conditions. Validation using real-world EV operational data demonstrates the effectiveness of the proposed framework. The proposed strategy provides an effective and interpretable solution for battery consistency evaluation and safety warning in EVs.