Spatio-temporal forecasting holds significant research value across numerous application domains, with traffic flow forecasting standing out as a typical representative due to its complex long-range spatio-temporal correlations, making it a highly challenging task. Existing methods typically model spatial and temporal dependencies separately using shallow graph convolutional networks and temporal feature extraction modules, but they often lack the ability to distinguish between isomorphic graphs. Furthermore, current approaches overlook the integration of spatial connections and temporal dependency features, which is crucial for comprehensively modeling traffic networks.To overcome these limitations, a novel Graph Isomorphic Neural Ordinary Differential Equation Network (GINODE) is proposed to capture the complex spatio-temporal dynamic relationships. By introducing Graph Isomorphism Networks (GIN), the model achieves the construction of deep networks and the synchronous modeling of spatio-temporal features. Experimental results demonstrate that the proposed GINODE model outperforms state-of-the-art baseline methods in terms of performance.