Accurate　traffic　forecasting　is　essential　for　smart　city　development.　However,　existing　spatiotemporal　modeling　methods　often　face　significant　challenges,　including　limitations　in　handling　complex　temporal　dependencies,　capturing　multiscale　spatial　relationships,　and　modeling　the　interaction　between　temporal　and　spatial　features.　These　challenges　arise　due　to　the　reliance　on　extended　historical　data,　fixed　adjacency　matrices,　and　the　lack　of　dynamic　spatiotemporal　interaction　modeling.　To　address　these　issues,　we　propose　the　Spatiotemporal　Dual-Scale　Interaction　Network　(SDSINet).　SDSINet　introduces　an　implicit　temporal　information　enhancement　method　that　embeds　temporal　identity　information　into　feature　representations,　reducing　the　computational　overhead　and　improving　the　modeling　of　global　temporal　features.　Additionally,　SDSINet　integrates　a　dynamic　multiscale　spatial　modeling　approach　that　combines　adaptive　and　scale-specific　graphs,　enabling　the　model　to　capture　both　local　and　global　spatial　dependencies.　Furthermore,　SDSINet　incorporates　a　dual-scale　spatiotemporal　interaction　learning　framework　that　captures　short-term　and　long-term　temporal　dependencies　as　well　as　multiscale　spatial　correlations.　Extensive　experiments　on　real-world　datasets　-　traffic　flow　(PeMS04),　speed　(PeMSD7(M)),　and　demand　(NYCBike　Drop-off/Pick-up)　-　demonstrate　that　SDSINet　outperforms　existing　state-of-the-art　methods　in　prediction　accuracy　and　computational　efficiency.　Notably,　SDSINet　achieves　a　14.03%　reduction　in　MAE　on　the　NYCBike　Drop-off　dataset　compared　to　AFDGCN,　setting　anew　benchmark　for　traffic　forecasting.