FD-Transformer: Fourier-Guided Long-Term Modeling for Lithium-Ion Battery RUL Prediction

Accurate prediction of the RUL of lithium-ion batteries play a crucial role in ensuring safety and reliability in energy storage systems. We introduce a novel Transformer-based RUL estimation framework that explicitly models the quasi-periodic degradation behavior of battery capacity. To improve the reliability of input feature representations, a Fourier-based denoising network is introduced, which efficiently captures periodic patterns through Fourier series expansion. Leveraging the Transformer architecture further improves long-range dependency modeling in the time-series degradation data. To address the high computational cost typically associated with Transformers, we replace the standard normalization layers with a lightweight alternative, significantly improving efficiency without sacrificing accuracy. Extensive experiments on public battery datasets demonstrate that our FD-Transformer model surpasses traditional baselines in predictive precision and in the speed of inference, offering a viable approach for real-time battery diagnostics.