FDNet: Remaining Useful Life Prediction for Lithium-Ion Batteries Based on Fourier Series and DenseNet

To enhance the prediction accuracy of the remaining useful life (RUL) of lithium-ion batteries, this study proposes a novel RUL prediction model, termed FDNet, which integrates Discrete Fourier Series (DFS) and Densely Connected Convolutional Networks (DenseNet). In contrast to traditional methods that rely on one-dimensional (1D) numerical representations of time-series data, this model utilizes a two-dimensional (2D) visual representation for direct RUL prediction. First, the DFS is employed to convert 1D time-series signals into a 2D time-frequency domain, enabling the extraction of high-dimensional features from the original data. Then, these 2D images are subsequently input into DenseNet, which learns the mapping between the image features and the RUL targets. Finally, the proposed model is validated using the NASA lithium-ion battery aging dataset, and the results indicate that FDNet achieves superior prediction accuracy and exhibits notable robustness.