Maximizing the performance of data-driven capacity estimation for lithium-ion battery

Predicting the cycle life of lithium-ion batteries is crucial for optimizing their long-term utilization. Traditional prediction methods struggle with real-world data due to diverse aging conditions and computational constraints. Our study circumvents these obstacles, exclusively utilizing discharge capacity curves. We introduce a data-driven method that refines similarity-based estimates using correction factors from neural network. Impressively, this approach, using just 300 cycles of discharge capacity curves, reduces error rates to 7.16% and 21.88% for primary and secondary test data, respectively, marking a significant reduction compared to the 27% and 46% errors reported by Severson et al. This study promises substantial potential for diverse applications requiring remaining useful life predictions. Our code is open-sourced and available at https://github.com/HyosikMoon/RUL-code/. Copyright (c) 2024 The Authors.