Artificial Intelligence and Machine Learning
Page 637 of 768Sources and links on page 637 (1)
Links embedded in the original report PDF, including citations and other references. These destinations have not been independently reviewed. Links open in a new tab.
Text on page 637
Fast Data Augmentation to Improve Degradation Trajectory Prediction Data-driven degradation prediction models can suffer from the inadequacy of high-quality training data. Li et al (2025) mitigated this issue by generating capacity fade curves synthetically using 3 parameters: Offset (governs cell-to-cell manufacturing variability), Slope (related to the rate of capacity fade) and Elongation (determines the occurrence of the “capacity knee” and the EOL). Training datasets were constructed by: (a) augmenting, and (b) partially replacing seed data (from the lab and field) with synthetic data. The viability of this method was demonstrated by successfully predicting the trajectory with deep and traditional ML models - CNN and GPR, respectively, though its performance in higher dimensional feature spaces remains to be seen.
Cell-testing efforts can be reduced by at least 50% by substituting synthetic data into an existing dataset. The addition of synthetic data showed an increase in prediction accuracy, while the partial replacement of real data with synthetic data led to similar model performances using limited data. Source: W. Li et al., Energy and AI (2025)
Extracted from the PDF; charts and images may read imperfectly.
More in Academia (12)
Page 637 · 6/6