ПЕРСПЕКТИВЫ ПРИМЕНЕНИЯ ГРАФОВЫХ СВЕРТОЧНЫХ НЕЙРОННЫХ СЕТЕЙ В ЗАДАЧАХ ПРОСТРАНСТВЕННОЙ КЛАСТЕРИЗАЦИИ ТЕРРИТОРИЙ

Abstract

The paper considers the actual problem of formation of interregional clusters using spatial data analysis and models based on graph neural networks. The aim is to develop the theoretical and methodological foundations of the application of graph models and deep learning methods for the study of interregional and intermunicipal relationships in the problems of building predictive models of the economic potential of territories. The paper shows the possibility of adapting the typical architecture of spectral graph convolutional network for object clustering. Theoretical foundations for the application of a spectral graph convolutional network for spatial clustering of territories have been formulated and mathematical modeling by the example of a random graph using the tools of opensource libraries networkx, numpy of Python has been performed. The developed approaches are promising for further developments in the construction of recommendation systems in the field of interregional cooperation due to the possibility of taking into account spatial data, as well as socioeconomic indicators of territories in a single model based on the study of network structures.