The increasing use of electric vehicles amplifies the demand for affordable charging infrastructure. By smart charging applications, operators of large-scale facilities of AC chargers can save costs on installation and lighten the load on distribution grids by avoiding high peaks and unbalanced loads. In the present paper, we consider the problem of phase-balancing in the context of non-ideal charging characteristics: some electric vehicles represent unbalanced loads to the grid, and some react to inputs in an unexpected nonlinear fashion. Furthermore, users expect a fair distribution of the limited charging power. In this light, we formally characterize fairness, choose to control load in real time and model smart charging as a time-discrete knapsack problem. In order to guarantee phase symmetry and increase charging efficiency, we develop a real current measurement filter and use it to solve the problem using a branch-and-bound algorithm and to approximate solutions with a greedy algorithm. We compare these solutions in representative simulations based on real charging data. Additionally, we evaluate the greedy algorithm on real charging infrastructure with up to 100 charging points. We conclude from the results that the greedy algorithm using measurements of charging behavior guarantees capacity and symmetry constraints and demonstrates comparatively adequate fair charging efficiency and applicability to computation on resource-constrained hardware.
This paper presents a high-resolution, publicly available measurement data set for IEC 61851 controlled AC charging of 16 different Electric Vehicles (EVs) from different manufacturers, comprising over 100 complete charging sessions. The data set includes phase currents, voltages, active power, and reactive power sampled at 50 Hz, enabling precise capture of transient phenomena such as overshoots and settling dynamics during setpoint transitions. Additionally, raw voltage and current waveforms are recorded at 5 kHz to reveal inverter behavior and power electronics characteristics relevant for grid impact assessment. Using these measurements, the study characterizes charging behavior and quantifies controllability in terms of delay, rise, and fall dynamics following setpoint changes, as well as steady-state compliance with IEC-defined current limits. Our results show substantial deviations from commanded setpoints, with certain EVs exhibiting steady-state errors up to 10 %, demonstrating that a single generic EV model is insufficient for accurate representation of heterogeneous EV fleets in optimization and aggregation applications. Therefore, we compare different charging models to the measurements and identify which model best represents each EV, providing guidance for selecting an appropriate charging model for a given application.