To address the problem of frequent errors and changes in the topology relationship of low-voltage distribution networks that reduce the accuracy of topology identification，a topology identification method based on manifold density peak clustering is proposed.The method consists of two stages：feature extraction and feature clustering.In the feature extraction stage，a manifold learning algorithm reduces feature redundancy and preserves the arbitrary shape distribution features of voltage data by manifold learning algorithm of extracting low-dimensional embeddings.In the feature clustering stage，based on low-dimensional manifolds of voltage data，the density peak clustering algorithm groups similar low-dimensional manifolds into the same cluster and separates different ones into distinct clusters，achieving topology identification.Experimental results have demonstrated that compared with the original density peak clustering algorithm，the proposed algorithm can identify clustering centers more accurately，improving the accuracy of low-voltage distribution network topology identification.