BALTAS, NICHOLAS-GREGORY, LAI, NGOC BAO, MARÍN ARÉVALO, LEONARDO VIDAL, Tarraso, Andres , RODRÍGUEZ CORTÉS, PEDRO
No
Ieee Energy Conversion Congress And Exposition
Proceedings Paper
Científica
01/01/2020
000645593602051
The presence of power electronics in today's power systems strengthens due to the wider integration of renewable energy and energy storage systems. Subsequently, the dynamical response becomes harder to model and understand. As a possible solution, coherency identification, among other applications, can reduce complexity. However, conventional tools possess limitations related to the assumptions need to be taken beforehand. In this paper, we propose a fully unsupervised variation of neural networks called the growing self organising maps (GSOM). The main advantage of GSOM over traditional methods is that network structure is not fixed, thus previous assumptions about the number of coherent groups or data structure are not necessary. A spreading factor controls the growth rate of the network allowing the analyst to choose the level of granularity whilst ensuring topology preservation. The effectiveness of the proposed algorithm is tested on the Nordic 32 power system.