Current Distribution Estimation of Parallel-Connected Batteries for Inconsistency Diagnosis Using Long Short-Term Memory Networks

In electric vehicle applications, lithium-ion batteries are usually used in parallel connections to meet the power and energy requirements. However, the impedance and capacity inconsistencies among the parallel-connected batteries (P-LiBs) can lead to uneven current distribution, resulting in accelerated aging and safety issues. Since it is impractical to equip current sensors for all battery cells, this work aims to estimate the uneven current distribution without additional hardware which can be used for inconsistency diagnosis. The characteristics of P-LiBs under inconsistency are investigated by experimental study, the current distribution, and voltage curve of P-LiBs that are found to exhibit different features under various inconsistency conditions. Consequently, a recurrent neural network (RNN) with long short term memory (LSTM) is adopted to estimate the current distribution using only the terminal voltage and total current information. The proposed method is validated with two parallel-connected cells and the experimental results indicate a good estimation accuracy in both inconsistent impedance and aging conditions. Furthermore, in the case of more cells in parallel, the trend and abnormal rise of branch currents are still accurately tracked in three- and four-parallel connection situations. Based on the estimated current distribution, the inconsistency faults within P-LiBs can be efficiently diagnosed.