Enhancing Battery Voltage Prediction with Deep Learning: A Comparative Analysis of LSTM and Traditional Models

The growing demand for efficient energy storage solutions has sparked increased interest in precise battery voltage prediction. In this study the Long Short-Term Memory networks, are applied in the time series forecasting, to improve battery voltage prediction compared to conventional models. The developed forecasting system includes three main parts: Pre-processing, modeling, and evaluation. In the pre-processing, the voltage and current time series are normalized, and divided to the test and train subsets. Then the windows of consecutive samples are generated for both train and test subsets. The forecasting LSTM models are learnt from training data set, in the modeling phase. Finally, the Mean Absolute Error (MAE) is used as the evaluation criterion. This model is compared to two other neural network models. The study concludes that LSTM outperforms the other models, highlighted by a significantly lower MAE. With a comprehensive methodology and successful experimental framework, this paper demonstrates the efficacy of LSTM in predicting battery voltage, indicating its superiority over simple neural networks for time series forecasting.