Fault Diagnosis of Lithium Battery Packs Based on Hybrid Attention-Enhanced CNN-GRU Model

Traditional methods for fault diagnosis in lithium batteries are characterized by high complexity and low recognition rates in the presence of multiple faults, often resulting in false alarms. This paper introduces an enhanced CNN-GRU model incorporating a hybrid attention mechanism. By combining the local feature extraction capability of 1D Convolutional Neural Networks (1D-CNN), the temporal modeling advantages of Gated Recurrent Units (GRU), and the dynamic focusing of critical features through a channel-space hybrid attention mechanism, the proposed model significantly improves the accuracy of identifying complex fault patterns such as short-circuit and sensor failures. When evaluated on battery-specific experimental data, the novel architecture attained 4.21% greater prediction precision than standard CNN implementations (98.91% vs 94.70%) and 3.38% improvement over CNN-GRU combinations (98.91% vs 95.53%). This improvement markedly reduces misclassification occurrences, providing a reliable solution for ensuring the safe operation of lithium batteries.