The invention discloses a relay protection device clustering method based on unsupervised learning. The relay protection device clustering method is characterized in that on the premise of no samples or few samples, a relay protection device is clustered through the fixed values of the device; the method comprises the step of relay protection device data preprocessing, the step of device clustering of unsupervised learning and the step of visualization, a replay protection device data processing module ensures that initialization device types needed by a device clustering module of the unsupervised learning are provided under the condition of no samples or few samples, the device clustering module of the unsupervised learning carries out type division on the device fixed values with high dimensional data and uncertain dimensions in the mode of statistical learning, a clustering result is shown by a visualization module in a tree mode, and classification optimization is carried out by absorbing expert knowledge. The method overcomes the difficulty of high-dimensional few sample learning and classifies mass relay protection devices in a production system, meanwhile, the feedback of the expert knowledge is utilized for solving the problem that the mass relay protection devices are difficult to classify, precise automatic clustering of the devices is achieved, and the precision and reliability of transmission of grid fault data in a power secondary system are improved.