Modeling and Simulation of Complex Systems Based on Graph Neural Networks

Modeling and simulation of complex systems are critical issues for understanding their structural and functional properties. The ability of graph neural networks (GNNs) to learn and represent the internal correlations within data provides a new approach for modeling and simulating complex systems. Currently, there are various types of GNN models involving frequency-domain, spatial-domain, generative, heterogeneous, and spatio-temporal models. These models are widely applied in complex system modeling and simulation research in multiple fields such as industrial internet, social networks, and supply chains based on specific tasks and scenarios. Starting from three representative tasks: network topology representation, dynamic evolution modeling, and system behavior and structure generation, this paper systematically analyzed GNN applications in modeling and simulation of complex systems, and its advantages in capturing multi-level structures, dynamic interactions, and nonlinear relationships within complex systems were explored. It provided a new reference for the research on the integration of complex systems with GNNs.