Based on the error analysis of the BDS(BeiDou navigation satellite system) triple-frequency carrier phase observations, the screening criteria for the optimal carrier phase linear combination coefficients was determined. For high-dimensional multifrequency mixed data sets, a weighted fuzzy C-means clustering algorithm was used to assign partial BDS triple-frequency carrier phase observations obtained by traditional ergodic search methods through assigning different weight values to different clusters on the same dimension. The combined observations were optimized for classification and selection, which effectively solved the problem of low efficiency of the traditional GNSS(global navigation satellite system) carrier phase observations selection method, and provided an idea for the optimal selection of the combined observation value coefficients of multi-system multi-frequency data. Finally, the classification results were analyzed, and the applicable range of all kinds of combined observations was determined. The integer ambiguity of the optimal combination is calculated by using the geometric-freed CIR(cascading integer resolution) algorithm and the measured data, and the feasibility and reliability of the method are proved.