Author(s) :

Jayden Yang.

Volume/Issue :

Abstract :

Lithium-ion batteries (LIBs) have gradually developed into a cornerstone of modern renewable energy systems over the past few decades. However, research efforts into LIBs have been severely limited by the lack of public LIB data, creating a developmental bottleneck in this field. In order to address this prominent global issue, this study develops and compares two contrasting synthetic generators, physics-aware (PADS) and distributional (DDS), to determine each method’s feasibility in real-world LIB research. Using both methods, the study first synthetically expands an existing public LIB dataset. Each synthetically generated dataset is then statistically compared to the original data through a custom multilevel evaluation framework to assess which approach better preserves statistical fidelity. Ultimately, this synthetic method comparison study aims to outline a superior model that can be utilized for data expansion in real-world LIB research. When writing this paper, it was initially hypothesized that the distributional synthetic data generation method would better preserve statistical fidelity than the physicsaware synthetic data generation method. Indeed, after cycling through the research process for both synthetic data generation methods, the results were generally consistent with the initial claims. However, deeper analysis into the synthetic data quality revealed instability in the regression relationships and substantial diagnostic limitations in both models, highlighting a significant limitation in their application. This study concludes that while distributional data generators are more well-suited for statistical fidelity and general LIB research, the utilization of each model relies heavily on its intended application. Therefore, the results of this research can be used as a foundation for further comparative investigations between synthetic data generation methods in the LIB research field.