Accurate prediction of lithium-ion battery state of health (SOH) is crucial for battery management systems. Existing deep learning-based prediction methods primarily focus on single temporal scale features, making it challenging to simultaneously capture multi-scale dynamic characteristics during battery degradation processes. To address this issue, a SOH prediction model based on multi-scale attention mechanism (MSANet) is proposed. The model employs three parallel attention modules - short-term, medium-term, and long-term - to capture features at different temporal scales, achieving comprehensive battery state modeling through an adaptive feature fusion strategy. Additionally, bidirectional LSTM (long short-term memory) and low-rank adaptation (LoRA) techniques are incorporated to enhance the model's feature extraction capability and parameter efficiency. Experiments on the NASA (National Aeronautics and Space Administration) battery dataset demonstrate that this method achieves superior prediction accuracy and efficiency compared to LSTM-based prediction methods, providing a novel solution for battery health state assessment.

lithium-ion battery  /  state of health prediction  /  multi-scale attention mechanism  /  time series prediction  /  low-rank adaptation