Hyperparameter Selection for the LSTM Model using in Electric Vehicle Battery State of Charge Estimation
Year: 2026
Type: conference-paper
Abstract: In electric vehicles (EVs), the state of charge (SOC) is crucial for the reliability of battery management systems (BMSs), energy management, and extending battery life. Recently, long short-term memory (LSTM) based deep learning (DL) models have been frequently used for SOC estimation. A review of the literature reveals that while studies have focused on model architecture development, hybrid deep learning approaches, and optimization algorithms, the impact of the fundamental hyperparameters of the LSTM model on estimation performance has not been sufficiently investigated. This study examines the effect of window size, batch size, and epoch number on SOC estimation performance in an LSTM model. A real field battery dataset was used in the study. After data preprocessing and normalization steps, the dataset was divided into 80% training and 20% testing. During hyperparameter analysis, only one hyperparameter was changed each time the model was run, while the other hyperparameters were kept constant. Model performance was evaluated using mean square error (MSE), mean absolute error (MAE), and root mean square error (RMSE) metrics. The results show that hyperparameter selection significantly affects the learning success and SOC estimation accuracy of the LSTM model. The most successful results were found at 72 for window size, 32 for batch size, and 50 for epoch number. In conclusion, it is shown that appropriate hyperparameter selection can significantly improve SOC estimation performance without any changes to the model architecture.
Institutions: Necmettin Erbakan University, Aksaray University
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7223710100