Abstract: Accurate acquisition of power supply phase sequence information in low-voltage distribution-station area plays an important supporting role in the development of lean management of line loss and accurate treatment of three-phase imbalance.Although the phase sequence of the three-phase meter has been gained during data collection, it is inconsistent with the real phase sequence due to non-standard wiring and other problems. At the same time, ignoring the three-phase meters will degrade the identification accuracy of the phase sequence. Therefore, this paper proposes a phase identification method for users in low-voltage distribution-station area based on feature constraint clustering of three-phase meters. Firstly, Z-Score and t-distributed stochastic neighbor embedding（t-SNE） algorithms are used to standardize and reduce the dimension of user voltage-time-series characteristics. On this basis, a constrained fast K-Medoids（CFK-Medoids） semi-supervised clustering algorithm based on background knowledge is proposed to cluster users. Finally, the actual data of station area in Guangdong Province of China are selected for case study. The results show that the proposed method has high identification accuracy and can effectively identify the problem of non-standard wiring of three-phase meters.