A Hybrid Transformer-BiLSTM Model for Lithium-Ion Battery Health State Prediction

This study proposes a novel deep learning model based on a hybrid Transformer-bidirectional long short-term memory (BiLSTM) architecture, which can achieve high-precision prediction of battery state of health (SOH) through a dual-attention mechanism (self-attention combined with feature-aware attention). The model innovatively integrates the global dependency modeling capability of the Transformer with the local temporal feature extraction advantages of BiLSTM. A cross-battery data fusion strategy is designed to enhance model generalizability. For the National Aeronautics and Space Administration (NASA) dataset, robust temporal samples are constructed using dynamic sliding window processing and Gaussian noise injection. Experimental results demonstrate that the model achieves state-of-the-art (SOTA) performance on the test set and exhibits early warning capability for remaining useful life (RUL) prediction. This work provides an interpretable and high-accuracy solution for predictive maintenance in industrial equipment.