The invention discloses a user clustering method and system based on a large-scale MIMO-NOMA system, and a medium, and the method comprises the following steps: 1, determining a channel matrix H according to a selected channel model, calculating the channel gain of a user, and carrying out the sorting; 2, determining boundary users and confirming an initial cluster head by calculating the channel coherence degree of the users; 3, calculating the channel coherence degree between the cluster head set and the candidate user set, setting a threshold value, selecting all cluster heads, and determining a cluster head set gamma; 4, calculating the channel coherence degree and channel difference between the remaining users and the cluster head set, setting a channel coherence degree threshold value delta 2, and finally determining the user grouping condition. The user clustering algorithm provided by the invention not only is beneficial to eliminating inter-cluster interference and increasing multiplexing gain, but also has better frequency efficiency performance compared with a traditional OMA structure under the same condition.