The proposal of "carbon peaking and carbon neutrality goals" target has brought revolutionary challenges to low-voltage distribution system. The novel low-voltage distribution network with clear topology and large-scale access of distributed optical storage system has been in the initial stage of construction. This paper analyzes the characteristics of the voltage spatial and temporal distribution of the novel low-voltage distribution network, and introduces a novel topology identification method for the low-voltage distribution network. The method typically extracted the most recognizable local feature sub-fragments (shaplets) of voltage curves without supervision, and formed u-shapelets distance graph as similarity measure for clustering, which could effectively improve the accuracy rate of topology recognition of conventional clustering algorithms. It avoids the problem that the identification accuracy decreases due to the narrowing of the distance between classes when clustering due to the overall flattening of the voltage spatio-temporal characteristics after the large-scale access of the optical storage system in the novel low-voltage distribution network. Moreover, it has high robustness to the non-fault absence of data scenarios.