In order to further improve the training effect of preschool education talents, this paper applies clustering algorithm to preschool education talent portraits based on the idea of big data analysis, combines the KD(K-Dimensional) tree to divide the data structure to accelerate the calculation, and uses the advantages of fast nearest neighbor node search and the advantages brought by Kmeans + + algorithm in selecting cluster centers according to distance to discuss the improvement of BI-Kmeans algorithm. Moreover, after constructing the preschool education talent portrait model, this paper analyzes and compares the computational complexity of the improved algorithms and the influence of cluster center selection on clustering results. Combined with the experimental analysis, it can be seen that the improved algorithm clustered SSE(Sum of Squared Errors) has been reduced to some extent and the clustering results tend to be stable. Through the analysis of SSE results, it is found that the improved algorithm has about 5% improvement in the sum of squares of errors. It is verified that the proposed portrait of preschool education talents has certain practical effect.