Cloud platform protocol data graph structure modeling method based on spectral clustering

Graph Convolutional Neural Network (GCN) used for protocol recognition has a high dependence on the input graph structure. Due to changes in proprietary protocols by manufacturers, it is difficult to establish an effective graph structure using traditional graph composition methods. To solve this problem, a graph structure construction method based on spectral clustering is proposed. Improve the similarity matrix of spectral clustering and choose Bray-Curtis distance to calculate similarity, use subspace learning and attribute selection to eliminate the influence of data redundant information, and use hypergraph to construct graph structure. At the same time, the hypergraph Laplacian matrix is used to constrain the spectral representation, and a graph structure model applied to GCN is established. Using protocol data in an industrial environment, the proposed spectral clustering algorithm is compared with other clustering algorithms. The experiments show that the improved composition method is accurate and efficient in high-dimensional data composition, and the new spectral clustering method reduces the result deviation.