An Informer-LSTM Network for State-of-Charge Estimation of Lithium-Ion Batteries

In this paper, we propose an Informer-LSTM hybrid model for lithium-ion battery state of charge (SOC) estimation. The Informer-LSTM model combines the strengths of the Informer model and Long short-term memory model to effectively capture the temporal dependencies and position features of the input data. By employing a sliding window mechanism, the long original data is divided into overlapping shorter segments, enabling the model to retain the relative time characteristics. The proposed model predicts multiple future SOC values at each time step, providing a comprehensive understanding of the battery's dynamic behavior. Extensive experiments are conducted on various charging/discharging modes and different temperature conditions. The results demonstrate that the model exhibits excellent generalization capability, with the majority of the tested data achieving root mean square error and mean absolute error of less than 1% in charging/discharging modes and temperatures not included in the training set. Furthermore, our model outperforms LSTM in terms of training speed, estimation accuracy, and generalization ability. Overall, our proposed model contributes to the advancement of SOC estimation and paves the way for realtime applications in practical settings.