Optimizing Electric Vehicle Workplace Charging Patterns Using Feature Selection and Machine Learning Models

Abstract

Electric vehicles are gradually gaining popularity, and organizations require a more efficient means of charging for their workplace usage and frequent, high usage. The present work analyzes a dataset of 3,395 EV charging sessions collected from a workplace charging program by the US Department of Energy (DOE). By performing detailed data analysis, we have identified session frequency, session length, and charging behaviors and applied and compared several machine learning models for future prediction. The bGA performance, analyzed with the bGGO and the bPSO, showed that the best average error, 0.53998, and the best fitness score, 0.59998, prove that bGA improves feature selection. When examining the overall performance of all the machine learning models used, MLPRegressor slightly outperforms them, with an R2 score of 0.8811, indicating high predictive accuracy regarding the characteristics of sessions. Calculating by presenting charging trends in weekly and daily bars or using radar plots helped to reconsider comparative model performance. Implications offer practical recommendations that address critical concerns listed in the literature, including directions for increasing workplace EV charging and strategies that promote the engagement and effectiveness of future workplace charging station efforts. In their field of EV management, this work provides a suitable framework for analyzing and modeling charging behaviors at the workplace.