State of Charge Estimation of Lithium Battery Based on Window Long Short-Term Memory Network

Accurate estimation of battery charge state is crucial for improving battery reliability and safety by preventing overcharge and overdischarge. This paper presents a simple and accurate neural network model, based on the window LSTM algorithm. The model uses sliding window to enhance data utilization and improve the learning of data relations. Cyclic charge and discharge experiments were carried out on the batteries with different C-rates. The model was trained and verified during the discharging process. The average absolute error was only 0.14%, and the root-mean-square error was as low as 0.0056. These results demonstrate the validity of the modeling approach.