Large-scale routing network expression method based on FCM

Abstract

The invention belongs to the field of network topology structure analysis, and particularly relates to a large-scale routing network expression method based on FCM. The method comprises the steps: carrying out the decomposition of a large-scale network topology structure through fuzzy mean clustering, and obtaining a sphere with coarsened granularity; carrying out random walk among all the spheresto obtain a macroscopic sequence, and carrying out random walk in the spheres to obtain a local sequence; finally, training the model based on an unsupervised learning algorithm SkipGram to acquire the low-dimensional vector expression of each device. According to the invention, based on the multi-granularity thought, an initial network is divided according to properties, three properties, namelydegree centrality, intermediate centrality and feature vector centrality, of network equipment need to be calculated, and efficiency and an improvement effect can be improved by considering from different granularities; therefore, compared with a traditional algorithm, the method is low in algorithm complexity, high in parallel capability and suitable for large-scale complex networks.

Description

technical field [0001] The invention belongs to the field of network topology analysis, and in particular relates to an FCM-based large-scale routing network expression method. Background technique [0002] With the development of satellite communications and the Internet, the scale of the network structure increases exponentially, and the analysis of complex large-scale networks has become an important issue at present. Traditionally, the adjacency matrix is ​​used to represent the network structure, and its dimension is the square of the number of devices n, which is unacceptable for large-scale networks in the real world. Moreover, most networks are sparse, and there will be a lot of redundant information represented by an adjacency matrix. Embedding-based network representation methods aim to learn continuous dense representations of devices in low-dimensional spaces, thereby reducing noise and redundant information. [0003] If the original network is directly trained...

Images