A recent proactive defense mechanism, named as moving-target-defense (MTD), has been proposed as a prevailing topic that is capable of actively changing transmission line reactance to preclude cyberattacks. However, the MTD strategy has seldom been studied for the unbalanced AC distribution system in the existing literature. Towards the end, this paper proposes a deeply-hidden MTD (DH-MTD) to elaborately hide both the self and mutual reactance of each phase at the transmission line installed with D-FACTS devices. Both the branch and injection power phasor measurement functions are integrated into DH-MTD in the cyberattack scenario and under the normal operating condition, while the system voltage stability is ensured. The proposed DH-MTD model is solved using a nonlinear least square (NLS) method based on the trust-region algorithm due to the non-Gaussian noise assumption. Also, we cope with the MTD allocation (MTDA) problem using a data-driven normalized PDF peak residual (NPPR) index. The effectiveness of the proposed DH-MTD method is demonstrated in the unbalanced IEEE 123-bus distribution system against both branch and node cyberattacks.