In order to analyze the multi-point data of high arch dams more efficiently, the kernel principal component analysis (KPCA) method for error reconstruction was introduced to identify the outliers in the multi-point sequence. To solve the problem of poor performance of the clustering by fast search and find of density peaks (CFSFDP) algorithm in high dimension, the KPCA method was used to reduce the dimension, and an improved CFSFDP (ICFSFDP) algorithm was proposed for automatic selection of cluster center and truncation distance. Based on KPCA-ICFSFDP method and the multiple output Gaussian process (MOGP) method, a multi-point deformation MOGP model was built according to the partition. The results show that for the small number of partitions within the cluster points, compared with the MOGP model without partition, the prediction effect is improved. Meanwhile, for the measurement points with good performance of the overall MOGP model, the prediction accuracy remains high after partitioning, which is improved compared with the single-output Gaussian process model.