Forecasting Electric Vehicles’ Charging Behavior at Charging Stations: A Data Science-Based Approach

• Published in: Energies (Basel) 2024-07, Vol.17 (14), p.3396
• Main Authors: Amezquita, Herbert, Guzman, Cindy P., Morais, Hugo
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Alternative energy sources; Automobiles, Electric; Battery chargers; Carbon; charging stations; Comparative analysis; Deep learning; Electric power systems; Electric vehicles; Electricity distribution; Emissions; Energy consumption; Engineering; EVs’ charging behavior forecasting; Feature selection; Forecasting; Information management; Machine learning; Neural networks; Renewable resources
• ISSN: 1996-1073; 1996-1073
• DOI: 10.3390/en17143396

The rising adoption of electric vehicles (EVs), driven by carbon neutrality goals, has prompted the need for accurate forecasting of EVs’ charging behavior. However, this task presents several challenges due to the dynamic nature of EVs’ usage patterns, including fluctuating demand and unpredictable charging durations. In response to these challenges and different from previous works, this paper presents a novel and holistic methodology for day-ahead forecasting of EVs’ plugged-in status and power consumption in charging stations (CSs). The proposed framework encompasses data analysis, pre-processing, feature engineering, feature selection, the use and comparison of diverse machine learning forecasting algorithms, and validation. A real-world dataset from a CS in Boulder City is employed to evaluate the framework’s effectiveness, and the results demonstrate its proficiency in predicting the EVs’ plugged-in status, with XGBoost’s classifier achieving remarkable accuracy with an F1-score of 0.97. Furthermore, an in-depth evaluation of six regression methods highlighted the supremacy of gradient boosting algorithms in forecasting the EVs’ power consumption, with LightGBM emerging as the most effective method due to its optimal balance between prediction accuracy with a 4.22% normalized root-mean-squared error (NRMSE) and computational efficiency with 5 s of execution time. The proposed framework equips power system operators with strategic tools to anticipate and adapt to the evolving EV landscape.