State of Health Estimation of Battery Pack for Real-World Electric Vehicles Based on User Behavior Profiling

Accurately assessing the state of health (SOH) for battery packs is crucial for effective lifecycle management of electric vehicles (EVs). Nevertheless, diverse user habits and complex operational data impose challenges. This study proposes a systematic SOH estimation framework that links macro‑level user behavior to micro‑level battery aging characteristics, enabling high‑precision assessment with low computational cost. User behavior profiles are built from normalized histograms of battery runtime and energy throughput across stress ranges. Multi‑level aging features with clear physical and statistical significance are derived from usage performance distributions and operational range trends across four hierarchical levels: vehicle, pack, cell, and inconsistency. Computationally efficient SOH estimation is achieved through feature selection and lightweight algorithms. Validation on real‑world data from hundreds of EVs yields an average absolute percentage error (MAPE) of 1.17% and root mean square error (RMSE) of 1.42% using only 25% of training data. Validations on two laboratory battery packs show a maximum error of 1.38%, confirming robustness. The framework exhibits scalability to EVs with varied battery configurations and operating conditions, and can be extended to health management of energy storage systems.