Predicting the Remaining Useful Life (RUL) of lithium-ion batteries is crucial for effective health management in electric and hybrid vehicles. The degradation processes of batteries are complex, non-linear behaviors influenced by a range of operational conditions, which poses significant challenges for accurate RUL prediction. Machine learning, particularly through data-driven approaches, has emerged as a powerful tool for predicting the RUL of lithium-ion batteries. However, traditional Transformer-based models, despite their potential, have encountered limitations in adequately capturing the complex dependency relationships among different battery features, which is crucial for accurate RUL estimation. To fill this gap, we propose a novel machine-learning model for RUL prediction. The model uses a Dimension-Slice-Wise Denoising Autoencoder(DSW-DAE) to preserve time and dimension information. In addition, the Two-Stage Attention(TSA) layer is proposed to capture the cross-time and cross-dimension dependency efficiently. We performed experiments and compared our method with existing approaches using the CALCE dataset. Comparative analysis against existing methods demonstrated that our model addresses the shortcomings of the previous Transformer-based approaches and shows superior performance in predicting the RUL of lithium-ion batteries.