Algorithm for link prediction in self-regulating network with adaptive topology based on graph theory and machine learning

The paper presents a graph model of the functioning of a network with adaptive topology, where the network nodes represent the vertices of the graph, and data exchange between the nodes is represented as edges. The dynamic nature of network interaction complicates the solution of the task of monitoring and controlling the functioning of a network with adaptive topology, which must be performed to ensure guaranteed correct network interaction. The importance of solving such a problem is justified by the creation of modern information and cyber-physical systems, which are based on networks with adaptive topology. The dynamic nature of links between nodes, on the one hand, allows to provide self-regulation of the network, on the other hand, significantly complicates the control over the network operation due to the impossibility of identifying a single pattern of network interaction.
On the basis of the developed model of network functioning with adaptive topology, a graph algorithm for link prediction is proposed, which is extended to the case of peer-to-peer networks. The algorithm is based on significant parameters of network nodes, characterizing both their physical characteristics (signal level, battery charge) and their characteristics as objects of network interaction (characteristics of centrality of graph nodes). Correctness and adequacy of the developed algorithm is confirmed by experimental results on modeling of a peer-to-peer network with adaptive topology and its self-regulation at removal of various nodes.