A reconstruction-based model with transformer and long short-term memory for internal short circuit detection in battery packs

Abstract

With the rapid growth of the electric vehicle industry, the demand for battery fault detection methods is also growing. Effective battery defect detection methods help maintain the performance of the battery pack. In this research, a reconstruction-based model for internal short circuit (ISC) detection in battery packs is presented by combining transformer and long short-term memory (LSTM). LSTM is added to the model to improve the encoding results of the transformer. The optimized decoders have a positive effect on the fitting ability of the model. In this model, the voltage data of the battery packs are taken as input and residual signals are generated by the differences between the reconstructed and true values to detect the ISC. The online parameter update method is proposed to increase the ability of the neural network model to screen out fault signals more efficiently, and the false-positive clipping method is employed to enhance detection accuracy. A large amount of data, including ISC, was collected from the laboratory to test the sensitivity and robustness. The verification results show that our model has better fitting ability and sensitivity. The parameter update method and false-positive clipping method help improve the adaptability and robustness. Our detection method achieves 96.05%, 98.57%, and 96.40% accuracy in ISC detection of 1 ohm, 5 ohms, and 10 ohms respectively.

Keywords

• Internal short circuit
• Reconstruction-based model
• Online parameter update method
• False-positive clipping method