Adaptive Multi-Criteria Optimization of EV Charging Decisions using XGboost and TOPSIS

Abstract

The paper presents user-specific multi-criteria analysis integrating predictive machine learning to develop a decision-making framework for electric vehicle charging optimization. The framework utilizes an XGBoost model trained on both synthetic and real-world data to predict optimal charger selections across various station-time combinations, with the goal of minimizing charging duration. Three user profiles — time-sensitive, price-sensitive, and speed-sensitive — are incorporated into a TOPSIS-based optimization, which utilizes predictions and includes charging price and speed. The results show that varying user types have significantly different preference for choices: each time-sensitive user chooses fast chargers day / time, each price-sensitive user chooses standard but less expensive chargers, and speed-sensitive users select chargers that provide the highest power charging rate independent of wait time and price. The limitations of static optimization are demonstrated, as well as value of a custom and adaptable approach. The model achieves high prediction accuracy and provides intelligent, user-centric recommendations regarding electric vehicle charging options.