A Cluster Comparative Learning Based Abnormal Detection Method for Power IoT Terminals

The widespread application of electric power IoT faces a large number of terminal devices sense access due to the complexity of terminal devices and differences in the threat of attack on the power grid is increasingly severe, in order to better ensure the security of the power IoT underlying terminals, this paper proposes a clustering-based comparative learning method for the detection of abnormal behavior of power IoT terminals. First, the sample data are extracted by the scalable self-organizing mapping network model. The mapping is completed to output the data with corresponding feature dimensions, and the data are downscaled and enhanced by the deep self-encoder. Then, it is input to the clustering comparison learning model (SwAV), and the positive and negative sample detection model is constructed by learning to group sample data with similar characteristics. Finally, the results of this model are input to the encoder for anomaly detection of unknown terminal data. The comparison experiments show that this scheme has good superiority and stability compared with other detection models.