However, their negative impact on the grid operation can be
mitigated by making them bidirectional, leveraging the energy stored in EV batteries or in the installed separate storage. Therefore, the power system can exploit this amount of energy to deal with unexpected grid large power imbalances. Moreover, ultra-fast dc chargers can contribute to power system stability by embedding virtual synchronous machine (VSM)
algorithms into their ac/dc stage, i.e., the active front-end (AFE) converter unit. The charging station is thus enabled to provide grid services normally in charge of traditional synchronous generators, such as inertial behavior and short circuit current injection during faults to trigger line protections. However, the provision of inertial active power involves a non-negligible reactive power contribution due to the active-reactive power coupling, thus increasing the output current of the converter. Nevertheless, the power coupling also affects the grid support during faults. Indeed, when the AFE injects a short circuit current into the grid, a fluctuating active power can propagate from the grid to the EVs, resulting in a potential cause of degradation for the EV batteries. Therefore, this article proposes a feedforward-based decoupling solution to
guarantee the complete active–reactive power dynamic decoupling while the AFE of an ultra-fast dc charger is providing grid support. Moreover, the proposed method ensures a full-decoupled dynamic response also in the case of power references variation during the normal EV charging operation. The proposed decoupling algorithm is experimentally validated on a down-
scaled 15 kVA two-level three-phase inverter, emulating the AFE of the ultra-fast dc charger.

Uncertain outage risk, collapse risk, and deicing workload of each power line are modeled as fuzzy values predicted by fuzzy deep learning models, and we transform the fuzzy optimization problem into a crisp optimization problem based on fuzzy arithmetics and uncertain theory. We propose an evolutionary algorithm, which combines global search without individual interaction and adaptive local search that uses a fuzzy inference system to determine the operator to be applied on each solution. The algorithm is fully parallelizable and therefore can solve the problem very efficiently based on GPU parallel acceleration. Computational results on real-world problem instances validate the performance of the proposed method compared to the state of the arts.