Optimized GRU‐Based Voltage Fault Prediction Method for Lithium‐Ion Battery Packs in Real‐Life

ABSTRACT Various failures of lithium‐ion batteries threaten the safety and performance of the battery system. Due to the insignificant anomalies and the nonlinear time‐varying properties of the cell, current methods for identifying the diverse faults in battery packs suffer from low accuracy and an inability to precisely determine the type of fault, a method has been proposed that utilizes the Random Forest algorithm (RF) to select key factors influencing voltage, optimizes model parameters through an Improved Dung Beetle Optimization algorithm (IDBO), employs a Gated Recurrent Unit (GRU) integrated with a channel and time attention mechanism (CTAM) for voltage fault prediction, and the consistency of the voltage is measured by quantifying the predicted voltage curve based on the curve Manhattan distance, a hybrid model for predicting voltage faults in lithium‐ion battery packs has been constructed, ultimately identifying faults such as overvoltage, undervoltage, and inconsistency of the battery pack. The experimental results show that the hybrid model proposed in this study outperforms the state‐of‐the‐art techniques such as informer and transformer in voltage fault prediction by achieving MAE, MSE, and MAPE metrics of 0.009272%, 0.000222%, and 0.246%, respectively, and maintains high efficiency in terms of the number of parameters and runtime.