Dariush Keihan Asl

This study addresses the problem of fair electric vehicle (EV) charging management in parking facilities, employing Jain’s Fairness Index (JFI) as the primary evaluation metric and, in some models, incorporating it directly into the objective function. The problem is formulated with practical constraints, including station power capacity, the number of active chargers, battery state-of-charge (SOC) dynamics, and distribution network limitations. Four scheduling policies are examined for comparison: First-In-First-Out (FIFO), Earliest Deadline First (EDF), Mixed-Integer Linear Programming (MILP) optimization, and reinforcement learning (RL) with a combined efficiency–fairness reward. Simulation scenarios are designed for 10, 50, and 100 EVs, as well as for varying proportions of fast-charging vehicles. Results indicate that the MILP approach achieves the highest level of fairness, while RL yields performance close to MILP by balancing efficiency and fairness. FIFO provides the lowest fairness level. An increased share of fast-charging vehicles causes a marked decline in the fairness index, whereas increasing the number of active chargers can lead to a more equitable energy allocation. The RL agent’s training process demonstrates gradual convergence of the average reward. For network constraint assessment, charging stations with capacities of 1,500 kW and 2,500 kW are modeled on standard distribution networks. Findings reveal significant sensitivity of power loss and voltage drop to the station’s connection location. Implementing the MILP policy, compared to uncontrolled charging, improves voltage stability and limits instantaneous power peaks. Overall, the proposed framework demonstrates that combining optimization techniques with reinforcement learning can substantially enhance fairness and user satisfaction while adhering to network constraints.