State-of-charge estimation of medium and high voltage batteries using LSTM neural networks optimized with genetic algorithms - CODE and DATA

The research focuses on predicting the battery’s state-of-charge by using historical SOC data as training input, comparing two predictive techniques: LSTM neural networks optimized with genetic algorithms (LSTM+GA) and multiple linear regression (MLR). The results show that the LSTM+GA model presents superior metrics (MSE: 0.0223; RMSE: 0.1494; MAE: 0.181) in comparison with the MLR model (MSE: 0.1590; RMSE: 0.3987; MAE: 0.2925), excelling in solving complex and nonlinear problems.