Simulating inter-city population flows based on graph neural networks

Abstract

Inter-city population mobility, a critical phenomenon in the modern urbanisation process, is closely related to urban industrial structure and socioeconomic development. This paper aims to investigate the dynamics of population flows and their intricate ties to industrial structure, so we employ the graph neural networks (GNNs) method to simulate inter-city population flows in China, which efficiently integrates demographic and socioeconomic data with Tencent migration big data while accounts for geographical relationships between cities. The results show that the model’s predictive accuracy using the CPC index was high for road and rail traffic and moderate for air transportation. A comparison with real-world data verified the model’s effectiveness in predicting the urban hierarchy and regional aggregation of flows. Using GNNExplainer, the results indicated that population size positively influenced population flow, while developed manufacturing reduced population mobility for road and rail traffic but increased it for air transportation. By conducting scenario simulations in Northeast China, we found that enhancing the region’s industry and consumer service industry could mitigate negative population outflows. The conclusions drawn from this study offer valuable perspectives to policymakers and urban planners, enabling them to make well-informed and judicious choices concerning urban planning, transportation, and resource allocation.