A region-aware field reliability prediction method for power batteries under heterogeneous conditions using improved deep clustering

Power batteries, as core components of new energy systems, exhibit significant regional variability in field reliability under heterogeneous operating conditions. However, accurate prediction of region-specific reliability remains challenging due to the scarcity of comprehensive field failure data across diverse regions and the homogeneity assumptions commonly adopted in existing regional reliability modeling. To address these limitations, this paper proposes a novel region-aware field reliability prediction method for power batteries under heterogeneous conditions, leveraging an improved deep clustering algorithm. First, environmental factors and a macro-level socioeconomic covariate associated with regional field reliability are systematically analyzed and identified, and a dynamic panel dataset is constructed. Mutual information is employed to quantify their statistical associations with regional field reliability. Second, an improved Transformer-based multi-kernel ensemble spectral clustering (TMESC) algorithm is developed to enable intelligent regional classification driven by heterogeneity in observed regional characteristics. Third, based on the derived regional clusters and available failure data from observed regions, a nonlinear mapping model is developed to quantitatively link field reliability with regional characteristics, facilitating accurate prediction of field reliability in unobserved regions. Experimental results show that, guided by TMESC clustering, the proposed method estimates power battery failure rates and has accurate predictive ability at the regional aggregate scale, improving regional reliability assessment and providing support for data-driven warranty and maintenance strategies.