In　the　construction　of　smart　cities　in　the　new　era,　traffic　prediction　is　an　important　component.　Precise　traffic　flow　prediction　faces　significant　challenges　due　to　spatial　heterogeneity,　dynamic　correlations,　and　uncertainty.　Most　existing　methods　typically　learn　from　a　single　spatial　or　temporal　perspective,　or　at　best　combine　the　two　in　a　limited　dual-perspective　manner,　which　limits　their　ability　to　capture　complex　spatio-temporal　relationships.　In　this　paper,　we　propose　a　novel　Multi-view　Spatio-Temporal　Dynamic　Fusion　Graph　Convolutional　Recurrent　Network　(MSTDFGRN)　to　address　these　limitations.　The　core　idea　is　to　learn　dynamic　spatial　dependencies　alongside　both　short-　and　long-term　temporal　patterns　through　multi-view　learning.　First,　we　introduce　a　multi-view　spatial　convolution　module　that　dynamically　fuses　static　and　adaptive　graphs　in　multiple　subspaces　to　learn　intrinsic　and　potential　spatial　dependencies　of　nodes.　Simultaneously,　in　the　temporal　view,　we　design　both　short-range　and　long-range　recurrent　networks　to　aggregate　spatial　domain　knowledge　of　nodes　at　multiple　granularities　and　capture　forward　and　backward　temporal　dependencies.　Furthermore,　we　design　a　spatiotemporal　attention　model　that　applies　an　attention　mechanism　to　each　node,　capturing　global　spatio-temporal　dependencies.　Comprehensive　experiments　on　four　real　traffic　flow　datasets　demonstrate　MSTDFGRN＇s　excellent　predictive　accuracy.　Specifically,　compared　to　the　Spatial-　Temporal　Graph　Attention　Gated　Recurrent　Transformer　Network　model,　our　method　improves　the　MAE　by　4.69%　on　the　PeMS08　dataset.