Efficient Neural Network-Based State-of-Charge Estimation for Electric Vehicles Through Transfer Learning

Electric vehicles (EVs) offer a green solution by minimizing environmental impact through zero tailpipe emissions and improved energy efficiency. Lithium batteries are crucial for EVs; however, accurately estimating their State of Charge (SOC) remains challenging. This study integrates neural networks with transfer learning to develop an efficient SOC estimation model that is data, resource, and time-efficient. Three distinct source datasets (LG, CALCE, and MADISON) are used to train seven pre-trained models, forming a diverse ensemble of hyperparameter-optimized neural networks. Seventeen transfer learning techniques are applied, resulting in 119 models evaluated on the PoliMi dataset using RMSE, MSE, and MAE metrics. Two scenarios are considered: one with abundant data, comparing a baseline model to transfer learning models, and a second with limited data, where transfer learning models significantly outperform, demonstrating their robustness in data-scarce conditions.