Rural population migration prediction method combining spatiotemporal graph convolution and social perception factors

The issue of rural population migration is closely tied to urban-rural structural adjustments and the effective implementation of the rural revitalization strategy. However, existing studies primarily focus on modeling structural geographic features, with limited consideration of the social factors that drive migration behavior. This leads to constraints in the explanatory power and policy adaptability of predictive models. To address this, this paper proposes a migration prediction method that integrates Spatio-Temporal Graph Convolutional Networks (STGCN) with socially-aware factors. A multi-source graph structure is constructed, incorporating structural connectivity, temporal dynamics, and social variables. Additionally, a graph attention mechanism is introduced to enhance the model’s ability to represent heterogeneous information. Simulation experiments conducted on a village-level migration dataset demonstrate that the proposed model outperforms mainstream methods such as LSTM and GCN in terms of average error and fitting accuracy. It also exhibits strong predictive sensitivity and interpretability under simulated policy intervention scenarios. The findings indicate that incorporating socially-aware factors significantly enhances the model’s responsiveness to migration intentions and policy variables, providing theoretical support and practical tools for regional population forecasting and rural governance.