To achieve accurate long-term traffic flow forecasting, we propose the Spatio-Temporal Dynamic Graph Mamba Model. Spatially, STMGNN hierarchically combines static topological graph structures with adaptive adjacency matrices to jointly model both dynamic and static spatial interactions. Temporally, the Mamba mechanism enhances long-term dependency modeling via a selective state-space architecture, which processes time-windowed sequences to efficiently capture global temporal correlations while maintaining linear computational complexity. Experimental results demonstrate that STMGNN outperforms baseline approaches in modeling spatio-temporal relationships, significantly advances intelligent transportation systems by enabling high-precision forecasts, providing critical support for urban planning and congestion mitigation strategies.