The　aim　of　this　study　is　to　construct　an　urban　traffic　flow　prediction　and　route　planning　system　based　on　deep　learning　models　to　solve　the　growing　problem　of　urban　traffic　congestion.　By　analyzing　and　processing　historical　traffic　data　of　a　city　and　using　a　hybrid　model　combining　Long　Short-Term　Memory　Network　(LSTM)　and　Graph　Neural　Network　(GNN),　we　are　able　to　accurately　predict　traffic　conditions　at　a　certain　time　and　location　in　the　future.　To　achieve　this　goal,　we　first　preprocessed　the　traffic　data,　including　converting　dates　to　temporal　features　and　One-Hot　coding　of　intersection　locations.　Subsequently,　we　used　deep　learning　techniques　for　training,　during　which　an　Adam　optimizer　as　well　as　a　categorical　cross-entropy　loss　function　were　used　to　improve　the　accuracy　and　reliability　of　the　predictions.　On　this　basis,　this　study　also　develops　an　optimal　path　planning　function　based　on　the　prediction　results,　which　provides　the　optimal　traffic　jam　avoidance　routes　by　adjusting　the　weights　of　the　road　network　and　applying　Dijkstra＇s　algorithm　to　find　the　optimal　paths　from　the　start　to　the　end　points.　The　experimental　results　show　that　the　system　achieves　good　results　in　traffic　flow　prediction　and　path　planning　with　an　accuracy　rate　of　91%.　©　2025　SPIE.