transformer-based lithium battery overcharge-induced thermal runaway prediction model
Description
The unified loss function and training strategy facilitate efficient end-to-end optimization, addressing practical requirements for battery safety prediction[11]. Based on the previously described innovative architecture design, the core structure of the overcharge-induced thermal runaway prediction model consists of four sequentially executed modules: (1) a feature embedding module that maps raw features into high-dimensional representation space, (2) a temporal modeling module capturing sequential dependencies through Transformer layers and multi-scale convolutions, (3) a feature fusion module that integrates global and local feature information, and (4) a classification prediction module outputting the risk probability of lithium battery overcharge-induced thermal runaway.
Files
Steps to reproduce
The experiment used 9,795 sets of battery operation data, with 7,927 sets (80%) for training and 1,957 sets (20%) for validation[20]. After 50 training epochs, the results demonstrated stable performance on the validation set (as shown in Figure 3), where the validation loss converged to 0.0135 and the highest validation AUC reached 0.943. These metrics indicate excellent generalization capability and convergence stability. No significant overfitting was observed during training, characterized by smooth parameter updates and moderate convergence speed.