Abstract
Diagnosing voltage faults of lithium-ion batteries is a critical function in the battery management system. Accurate diagnosis of voltage faults is crucial for ensuring the safety and reliability of energy storage applications and electric vehicles (EVs). This article proposes an optimized Graphical Neural Network (GNN) model. Specifically, the optimized GNN model extracts the relationships between various batteries by learning the topology of the batteries. The proposed method combines the physical coupling between batteries and the entanglement of measurement results with the strong nonlinear processing capability of neural networks to improve the effectiveness of fault localization. Experimental results on three publicly available datasets show that the proposed method outperforms baseline methods such as GraphConv, GCNConv, ChebConv, SGConv, CNN, DBN, LSTM and CNN-LSTM in terms of Accuracy, Precision, Recall, and F1-score, which verifies the effectiveness and accuracy of the proposed method in fault localization of voltage data. Compared to the highest-performing baseline method, the proposed method achieves a maximum improvement of 4.31% and 3.68% in the accuracy of abrupt fault and gradual fault localization respectively. This indicates that the proposed optimized GNN method for diagnosing voltage faults has satisfactory accuracy and stability, which is of remarkable significance for the development of EVs.