融合LSTM与Transformer动力电池故障诊断算法研究OA

To achieve accurate warning of power battery faults,this paper proposes a new method that integrates LSTM and Transformer real-vehicle fault diagnosis algorithm.First,operational data from ternary lithium batteries on vehicles are collected and preprocessed.Then,the Pearson correlation coefficient method is employed to select features exhibiting high correlation with individual cell voltages.Next,three diagnostic algorithms(LSTM-Transformer,Transformer,LSTMand GNN)are introduced to predict cell voltages on both sound and faulty vehicles.Voltage discrepancies are obtained by calculating the differences between predicted values and actual measurements.Finally,the local outlier factor(LOF)is implemented to compute LOF values for these voltage discrepancies.Fault determination is accomplished through threshold configuration to identify abnormal battery cells.Results show LSTM-Transformer model has the optimal model evaluation metrics and no false alarms compared to the Transformer model,LSTM model,and GNN model on sound cars.On faulty vehicles,the proposed algorithm successfully identifies defective cells and predicts thermal runaway events 40 hours in advance,with no false alarms or missed detections.These findings verify the effectiveness of the proposed algorithm for power battery fault diagnosis.The research may provide some insights for implementing proactive maintenance strategies of battery management systems on electric vehicles.