A Novel Transformer Model with Trend-Aware Self-Attention for Accurate Lithium-Ion Battery Capacity Prediction

Accurately predicting the capacity of lithium-ion batteries is critical for ensuring the reliability and efficiency of modern energy storage systems. This paper proposes a novel Transformer-based model that incorporates a trend-aware self-attention mechanism to address the limitations of traditional Transformers in capturing local trends. The proposed model leverages 1D convolution operations within the attention mechanism to effectively extract both global dependencies and local patterns from battery performance data. By integrating Add & Norm layers and feedforward networks, the model achieves enhanced feature representation and computational stability. Experimental evaluations using the NASA lithium-ion battery dataset demonstrate that the model consistently outperforms existing state-of-the-art methods in prediction accuracy, as measured by relative error (RE), mean absolute error (MAE), and root mean square error (RMSE) metrics. While the model shows strong performance, future research will focus on addressing uncertainties in real-world applications to further enhance its robustness and practical usability.