Aiming　at　the　problem　that　the　traffic　flow　has　an　low　prediction　accuracy　due　to　varying　randomness　and　high　nonlinearity,　an　improved　TSK　fuzzy　neural　network　prediction　model　by　singular　spectrum　analysis　is　proposed.First,　the　original　traffic　flow　time　series　can　be　decomposed　into　several　subseries　by　singular　spectrum　analysis　(SSA).　Then,　the　Takagi-Sugeno-Kang　(TSK)　fuzzy　system　was　trained　using　simulated　annealing　genetic　algorithm　(SAGA)　and　minibatch　gradient　descent　method　(MBGD)　combined　with　three　techniques　(RDA)　of　regularization,　DropRule　and　AdaBound.The　prediction　model　of　TSK　fuzzy　neural　network　is　established　to　predict　each　subseries　independently,　and　the　forecasted　traffic　flow　is　obtained　by　superimposing　the　prediction　results　of　each　subseries.　Finally,　the　traffic　flow　dataset　of　expressway　tunnels　in　a　city　is　used　to　verify　the　effectiveness　of　the　model.　Experimental　results　can　show　high　traffic　flow　prediction　accuracy　of　the　model,　and　the　model　can　accurately　predict　multiple　different　highway　tunnels.　©　2023　IEEE.