Internal short circuit fault diagnosis based on 1DVCNN in lithium battery storage system

Because LiFePO4 batteries are widely used in the energy storage system, their safety has received a great deal of research. If an internal short circuit (ISC) in a Li-ion battery energy storage system leads to thermal runaway, it will pose an uncontrollable hazard. Therefore, there is a proposed method in this study that is a one-dimensional voltage-correlated convolutional neural network (1DVCNN) to detect and localize ISC faults in time, which is a combination of feature extraction, correlation analysis, and convolutional neural networks. First, the voltage signal of the LiFePO4 battery collected in the energy storage system is then converted into a Pearson Correlation Coefficient (PCC) by correlation analysis of the data. Specific PCCs are then selected as features of a one-dimensional convolutional neural network (1DCNN) to diagnose faults. Finally, the fault dataset is used to train the model and the results are then validated against the untrained fault data. In this paper, the ISC faults are modeled by both building an actual training platform for lithium batteries and MATLAB simulations to obtain the dataset used to train the proposed model. By comparing the accuracy of the training results of the models using different features, it is found that 1DVCNN is able to detect the ISC fault in the energy storage system in a more timely and effective manner than in other studies.