The country is built on the people, and the people regard food as their heaven. Food security is the foundation for the healthy and stable development of human society. Natural disasters, local wars, climate change, and other events have affected China’s food security and social life. Food security is one of the cornerstones of global economic and social development. The structural robustness of the international rice trade network plays an important role in the global economy. The graph neural network algorithm and utility function theory are integrated for learning a trade decision-making model which contains the benefit endowments and cost endowment of economies in international trades. How to address changes in the international rice environment and the impact of international emergencies is a question of important research value. By integrating graph neural networks, utility theory, and other methods, heterogeneous individual characteristic representations are learned from international rice trade network data, and the network formation and evolution mechanism are revealed in complex systems. Then, the evolution of complex networks is simulated at the macro and micro levels to study the international rice trade networks in depth. Precise and controllable trade strategies, considering the impact of the trade war and COVID-19, are proposed. In the international rice trade network, Asia and Europe, North America and South America, and Africa and Oceania are three groups with similar vulnerabilities. As trade relations increase, India and Pakistan are less affected. The rice trade of the mainland in China is more affected. The model framework in this paper is highly scalable and transferable. The main innovation of the model framework is to connect the two research paradigms of data-driven and mechanism modelling. It is a general model framework that can be applied to different complex systems and complex networks across different fields.