The randomness of electric vehicle (EV) charging has negative impacts on three‐phase imbalance and peak–valley difference in electric energy distribution systems. Traditional EV charging strategies have shortcomings: the performance of three‐phase imbalance mitigation may be limited if the grid‐connected EVs are extremely imbalanced on three phases; in addition, the comprehensive regulation of peak–valley difference and three‐phase imbalance is not developed, and the three‐phase imbalance of reactive power is ignored. Therefore, a real‐time multilevel energy management strategy (RMEMS) for EV charging is proposed. A tri‐level optimization model (TOM) is designed as the central system. In upper‐level optimization, the three‐phase selection (TPS) of EVs is optimized to balance active or reactive power consumption on three phases. Based on the results from upper‐level optimization, the charging active power is regulated in middle‐level optimization to reduce the peak–valley difference on each phase. In lower‐level optimization, the reactive power compensated by EV chargers is optimized based on the results from upper‐level and middle‐level optimization to balance the reactive power on three phases. Case studies show that the proposed RMEMS performs well for balancing active and reactive power consumption on three phases, and the peak–valley differences of active power consumption on each phase are all mitigated.