Unsupervised Strategies to Network Topology Reconfiguration Optimization with Limited Link Addition

Autor(es)

Paiva, William R. ; Martins, Paulo S. ; de Angelis, Andre F. ; Barbosa, H ; GomezGardenes, J ; Goncalves, B ; Mangioni, G ; Menezes, R ; Oliveira, M

• Tipo de documento: Artigo Científico
• Fonte: COMPLEX NETWORKS XI; v. N/A, p. 9-pg., 2020-01-01.

Resumo

The evolution of networks may lead to an undesired configuration of its properties due to different forces driving their growth. A planned topology change aiming to bring the properties to an acceptable range is called Network Topology Reconfiguration Optimization with Limited Link Addition (NTRLA). We faced an NTRLA problem when we were investigating ways to improve the efficiency of large power grids. In the search for solutions, we developed strategies to add new edges in unsupervised automatic applications. The strategies were tested over thousands of realizations of random and scale-free networks, as well as over a power grid map by means of computer programs that have implemented them and collected the efficiency of the networks along the change processes. An attempt to determine the maximum possible performance provided a comparison reference to the strategies. We show that the best result was obtained by linking the node with larger closeness to the one with the smallest closeness, although this procedure is not scalable. Furthermore, the strategy that links nodes whose betweenness values are near to the median was shown to be scalable and easy to implement, even if it has not delivered the best performance. We have found that the min-cut procedure has improved the results for each strategy. It became apparent how the network topology plays a fundamental role in NTRLA problems, what prompted for new insights and further research work in the field of complex networks. (AU)

• Processo FAPESP: 10/50646-6 - Aplicação de computação de alto desempenho em problemas interdisciplinares
• Beneficiário: Vitor Rafael Coluci
• Modalidade de apoio: Auxílio à Pesquisa - Regular