Aiming at the defects of K-means clustering algorithm that the number of clusters is unknown in advance and cannot deal with non-convex distributed data sets, a clustering algorithm based on evolutionary idea and its cluster fusion algorithm are proposed, The algorithm embeds the K-means clustering algorithm into the framework of evolutionary clustering algorithm. By adjusting the distance doubling parameter, the data objects will be divided gradually, and the number of clusters k will be determined adaptively, Then, a middle circle density cluster fusion algorithm based on the nearest distance and a middle circle density cluster fusion algorithm based on representative classes are proposed to fuse the clusters with high similarity, so that the k value gradually tends to the real value. Experiments show that this method has good practice.