A similarity-based predictive scheduling method for dynamic electric vehicle charging load management

Area of Science:

• Electrical Engineering
• Computer Science
• Sustainable Energy

Background:

• Managing energy demand from public electric vehicle (EV) charging stations presents challenges to distribution grid stability due to uncertain user demand.
• Classical methods often struggle with real-time adjustments and user-specific charging requirements.

Purpose of the Study:

• To develop a multi-stage, data-driven control framework for real-time energy management at public EV parking lots.
• To achieve load profile smoothing, ensure user-defined charge levels at departure, and maintain grid stability.

Main Methods:

• A three-layer algorithm integrating historical similarity-based prediction, dynamic predictive optimization using a genetic algorithm, and a final repair stage.
• Stochastic scenario evaluations to ensure operational robustness.

Main Results:

• Reduced peak-to-average ratio (PAR) through load shifting from peak to off-peak hours, demonstrating effective peak shaving and valley filling.
• Limited maximum load ramp rate, preventing transformer stress.
• Achieved a final state of charge (SoC) error below 0.1% for all EVs.
• 35% reduction in charging pile occupancy with average processing times of a few seconds.

Conclusions:

• The proposed framework offers a robust, real-time solution for EV charging energy management, enhancing network resilience.
• Effectively balances grid stability with user demand, postponing infrastructure upgrade needs.
• Demonstrates high convergence towards optimal solutions for energy management.