The rapid global increase in electric vehicles (EVs) usage is leading a significant burden on charging infrastructure and requires precise energy management solutions for grid stability. Accurate forecasting of daily energy consumption at charging stations is critical for minimizing operational costs and optimizing infrastructure planning. This study presents a comprehensive comparative performance analysis of the deployment of different learning models to predict the daily energy demand of EV charging stations. In this paper, predictions have been performed based on variables such as the number of vehicles, charging time and connection time using real-time charging data. Machine Learning (ML) algorithms such as the Decision Tree, Random Forest, Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Bayesian Ridge, Hist Gradient, XGBoost, LightGBM, CatBoost, Ensemble Learning and Deep Learning (DL)-based TabNet have been integrated to resolve the complex and nonlinear structure of the dataset. The prediction performances of the models have been compared using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and R-Squared (Coefficient of Determination-R²) metrics. According to the analysis results, the TabNet architecture has been the most successful model in terms of overall performance, achieving the lowest RMSE of 37.30 and the highest R2 score of 0.9724. It has been followed by the Ensemble Voting model, which shared the same coefficient of determination (R2: 0.9724) and produced the lowest MAE value of 28.77. The obtained testing results demonstrate that TabNet and Ensemble Voting outperform traditional methods in capturing complex user behaviors and variable charging patterns.
Teke et al. (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: