Allocating the Power Module for Future EV: a Constraints Learning Approach

Diving into one of the increasing load demands for modern power systems, the charging demand from ultra-fast electric vehicle charging hubs (UFEVCH) has faced the challenge of distributing appropriate charging power to arriving electrical vehicles (EVs) by allocating power modules (PMs) that support a specific charging demand. In UFEVCH, PMs are gathered into a power-sharing pool, while only partial PMs can be accessed by the dispenser connected to the EV. In the meantime, neither the dispenser selection nor the charging demand of the EV are unknown in advance. Therefore, the PM allocation program developed in this study operates online; that is, the program would operate once the EV is connected. Moreover, to ensure the experiences of EV users, an interval forecasting model based on a neural network is developed to forecast the possible charging power range for the next-arrived EV, which will be reserved to ensure charging of EV and integrated as a constraint. Training on 20461 practical EV charging data efficiently integrates the learned constraint into the program. Comparisons from scenarios derived from practical EV charging data reveal that PM allocation with learned constraints could efficiently increase the service quality of the UFEVCH.