Research on the spatiotemporal prediction of supply chain finance risk contagion based on a dynamic graph attention diffusion mechanism

This study focuses on the spatiotemporal prediction of supply chain finance risk contagion and constructs a model based on the dynamic graph attention diffusion mechanism for in-depth exploration. The risk-aware dynamic graph attention network (DyGAT) is used to model the time-varying network structure of supply chain finance, where the spatial attention weight α_ij(t) is updated adaptively at each time step by integrating real-time risk intensity signals to realize real spatiotemporal coupling modeling. A bounded diffusion mechanism with natural risk attenuation is introduced to characterize the propagation process of risks in spatial and temporal dimensions, avoiding unbounded cumulative defects in traditional formulas. Methodologically, a supply chain finance dynamic graph structure is constructed by integrating multisource data such as historical transaction data and corporate financial information, with clear data cleaning, fusion and feature engineering processes, and the model is trained, validated and ablated strictly. The results show that the model can effectively predict the spatiotemporal paths and topological characteristics of supply chain finance risk contagion and identify key nodes and critical periods of risk propagation, and the ablation experiment verifies that the dynamic attention module and bounded diffusion module bring 18.2% and 12.5% precision improvement, respectively. It provides strong support for financial institutions and supply chain enterprises to formulate risk prevention strategies in advance and is of great significance for improving the stability and resilience of the supply chain finance system. However, caution is needed when applying the model to small-sample settings, as the limited sample size may constrain its generalizability to other industries and regions.