Large-scale spectral survey projects such as LAMOST produce a great deal of valuable research data, and how to effectively analyze the data of this magnitude is a current research hotspot. Clustering algorithm is a kind of unsupervised machine learning algorithm, which makes the clustering algorithm deal with the data without knowledge of the domain, and internal law and structure will be found out. Stellar spectral clustering is a very important work in astronomical data processing. It mainly classifies the mass spectral survey data according to its physical and chemical properties. In this paper, we use a variety of clustering algorithms such as K-Means, Bisecting K-Means and OPTICS to do clustering analysis for the early M-type stellar data in LAMOST survey. The performance of these algorithms on the early M-type stellar data is also discussed. In this paper, the performance of the Euclidean distance, the Manhattan distance, the residual distribution distance for the three clustering algorithms are studied, and the clustering algorithm depends on the distance measurement algorithm. The experimental results show that: (1) The clustering algorithm can well analyze the spectral data of the early M-type dwarf star, and the cluster data produced by clustering is very good with the MK classification. (2) The performance of the three different clustering algorithms is different, and Bisecting K-Means has more advantages in stellar spectral subdivision. (3) In the cluster at the same time it will produce some small number of clusters, and some rare celestial bodies can be found from these clusters. OPTICS is relatively suitable for finding rare objects.