Lithium ion battery life prediction based on Transformer-BiGRU model

To address the limitations of traditional methods in time series feature extraction and long-term dependency modeling for lithium-ion battery remaining useful life (RUL) prediction, this paper proposes a hybrid deep learning model that integrates Transformer and bidirectional gated recurrent unit (BiGRU). The model adopts a dual-channel architecture: Transformer captures the global long-term dependencies of capacity degradation using multi-head self-attention, while BiGRU extracts local temporal features. Through a feature interaction gating mechanism, the model dynamically allocates attention weights to achieve adaptive adjustment of hidden layer outputs and multi-dimensional feature fusion. The model is validated on the NASA battery aging dataset and demonstrates significant improvement in RUL prediction accuracy.