Based on the clustering idea, two-dimensional and three-dimensional cloud phase clustering identification algorithms are designed. Combined with the data of polarization lidar, microwave radiometer, and millimeter wave radar, the fine identification and classification methods of cloud phase are studied. The volume depolarization ratio, temperature, and reflectivity factor of cloud particles are taken as the input characteristics of clustering network learning, and the cluster division results of distinguishing different cloud phase are obtained through unsupervised learning. The fine identification of cloud phase is realized by using the cluster division results, which solves the problem of misjudgment of results caused by the single threshold algorithm in traditional cloud phase identification. Using this algorithm, the supercooled water, warm cloud liquid water, and ice phase can be identified efficiently. At the same time, the ice and water dominated mixed phase can be classified in detail. The clouds over Xi’an are observed by the polarization lidar, microwave radiometer, and millimeter wave radar, and the synchronous observation data of three instruments are retrieved. The two-dimensional clustering algorithm and three-dimensional clustering algorithm are used to identify and analyze the cloud data observed from January 9 to 10, 2021 and June 8 to 9, 2021. The distinction of liquid water, mixed phase (ice dominated and water dominated), supercooled water, and ice phase in the cloud is realized. Through comparison and analysis, it is found that the three-dimensional clustering recognition algorithm can show more details of the phase transformation process than the two-dimensional clustering recognition algorithm. The overall identification results are consistent with the actual weather transformation process.