2000国家大地坐标系（CGCS2000）作为正式发布的法定坐标系已运用了多年。在我国，所有基于全球导航卫星系统（GNSS）测站的坐标为了与CGCS2000框架保持一致，都需要进行坐标改正。实现最佳CGCS2000框架需采用不同的策略，而不同的策略会导致不同的结果，有的差异甚至达到几分米。GNSS测站坐标改正常用的两种方法是CGCS2000控制下的拟稳平差和板块运动改正，两种方法计算的结果相差超过10 cm。本文将监督聚类（supervised clustering）统计方法应用于GNSS基准站的选择，同时提出了GNSS测站大网数据处理分组的间距分区（partition spacing）法，并用板块运动改正将当前历元GNSS测站坐标归算至CGCS2000参考历元。结果表明，新的分区方法明显优于传统的地理分区方法。当以不分组的测站坐标为标准时，新分区方法得到的三维坐标分量的精度均优于2 mm。监督聚类法得到的x、y、z方向上的速度均方根（RMS）分别为0.19 mm·a^–1、0.45 mm·a^–1和0.32 mm·a^–1，远小于传统方法的0.92 mm·a^–1、0.72 mm·a^–1和0.97 mm·a^–1。此外，采用奇异谱分析（SSA）对位置非线性运动进行建模和预测。在东、北、高（E、N和U）方向，SSA的建模精度分别优于3 mm、2 mm和5 mm，在水平方向和垂直方向的预测精度分别优于5 mm和1 cm。

The China Geodetic Coordinate System 2000 (CGCS2000) has been used for several years as a formal published reference frame. The coordinates of all global navigation satellite system (GNSS) stations in China need to be corrected to align with the CGCS2000 frame. Different strategies can be adopted for the realization of an optimal reference frame. However, different strategies lead to different results, with differences as great as several decimeters when GNSS station coordinates are transformed into CGCS2000-defined coordinates. The two common methods for the coordinate correction of a GNSS station are quasi-stable adjustment under CGCS2000 and plate movement correction, and the differences between their results can be greater than 10 cm. In this study, a statistic method called 'supervised clustering'is applied to the selection of GNSS reference stations; a new scheme named 'artition spacing' for the grouping of all processed GNSS stations is proposed; and the plate movement correction method is used to correct the coordinates of all GNSS stations from the GNSS epoch to the CGCS2000 epoch. The results from the new partitioning method were found to be significantly better than those from the conventional station-blocking approach. When coordinates from the stations without grouping were used as the standard, the accuracy of all the three-dimensional coordinate components from the new partitioning method was better than 2 mm. The root mean squares (RMSs) of the velocities in the x, y, and z directions resulting from the supervised clustering method were 0.19, 0.45, and 0.32 mm∙a⁻¹, respectively, which were much smaller than the values of 0.92, 0.72, and 0.97 mm∙a⁻¹ that resulted from the conventional approach. In addition, singular spectrum analysis (SSA) was used to model and predict the position nonlinear movements. The modeling accuracies of SSA were better than 3, 2, and 5 mm in the east (E), north (N), and up (U) directions, respectively; and its prediction accuracies were better than 5 mm and 1 cm for the horizontal and vertical domains, respectively.