부분 방전 구간에서의 배터리 열화 합성데이터 생성을 위한 물리-데이터 기반 하이브리드 프레임워크

초록

Battery degradation data acquired under restricted Depth-of-Discharge (DoD) conditions reflect real-world operating environments; however, constructing training datasets remains difficult because of high acquisition costs and limited distributional coverage. To address these limitations, this study proposes a physics–data hybrid synthetic data generation framework comprising normalization, physical parameter identification, physics-based seed generation, surrogate-model-based large-scale generation, residual correction in a summary space, and quality-control gates. The proposed framework evaluates both group-level distributional alignment and physical consistency using data acquired within a restricted State-of-Charge (SoC) range of 40%–85%, and includes only synthetic data satisfying the quality criteria in the final training dataset. In the internal experiments, 375,207 restricted-DoD synthetic samples were generated from six LG M50L cells and 4,027 measured cycle summaries. The framework was also applied progressively to external public datasets to assess its broader applicability.

키워드