With a large number of power distribution substations (transformers), three-phase imbalance is a commonly observed issue. Due to the high randomness of single-phase loads in the substation area, the amplitude ranking of three-phase voltages and the imbalance degree according to the traditional algorithm change constantly. Based on the feature analysis of the sampled data, differences in the dominant phase amplitudes of three-phase voltages in each substation area during a certain period of time are obtained, identifying the severely unbalanced substation area for manual intervention. This serves as a feasible method to balance costs and benefits. A time-series matrix of the three-phase voltages for each substation area is constructed in this paper, and a method for analyzing the dominant features of phase amplitude differences based on singular value decomposition (SVD) and constructing related indexes is proposed. The results show that the submatrix corresponding to the maximum singular value is the absolute dominant component of the original three-phase voltage time-series matrix. The ranking of the dominant phase amplitudes can be obtained from the left singular vector, and the three-phase dominant imbalance index can be constructed using the maximum singular value and the left singular vector. This method addresses the problem of constantly changing phase amplitude ranking and can determine the amplitude of each phase for any selected time period. A test case verifies the dominance and uniqueness of the proposed imbalance indexes. Based on the proposed indexes, through a rolling cycle of identification and manual intervention, the overall three-phase balance of the distribution network can be improved and maintained. The findings offer valuable practical guidance.