Enhancing data-driven-based state of health estimation for diverse battery applications through effective feature construction

Data-driven-based State of Health (SOH) estimation methods are popular for their ability to avoid considering the battery's complex aging mechanisms. However, aging features extracted from specific voltage intervals and the SOH estimation models established under specific conditions may no longer be applicable to other battery applications. This work first reveals the necessary conditions for constructing effective aging features in a partial voltage interval, and specifically proposes an applicable method for determining the voltage interval of constructing aging features based on the incremental capacity curve and the voltage curve's shape. The proposed method can be adapted to different battery applications and is suitable for partial charging conditions. Given the different aging patterns present in different battery types and applications, a feature self-scaling -based online learning strategy is developed to reduce the model's reliance on aging datasets and enable it to adapt to real-time battery aging patterns. The proposed methods are validated through both an actual aging experiment and three publicly available aging datasets, showcasing their effectiveness in extending the applicability of the proposed SOH estimation methods for diverse battery applications.