Aiming　at　addressing　the　dynamic　response　problem　of　spatiotemporal　multiscale　pipeline　vibration　with　high-　and　low-frequency　superposition,　a　Fourier　Feature　Physics-Informed　Neural　Network　(FF-PINN)　is　proposed.　To　achieve　this,　the　Fourier　feature　mapping　is　introduced　to　facilitate　the　decomposition　of　temporal-　and　spatial-scaled　information.　By　employing　the　hyperparameter　σ　to　regulate　the　neural　network＇　s　learning　frequency　range,　the　comparatively　low-frequencies　at　the　macroscopic　time　scales　and　the　comparatively　high-frequencies　at　the　microscopic　scales　of　the　vibration　response　of　the　pipeline　can　be　effectively　captured,　so　the　spectral　bias　of　PINN　in　learning　high-frequency　information　may　be　obviated.　By　applying　FF-PINN　to　simulating　the　vibration　characteristics　of　the　pipeline　conveying　fluid　simply　supported　at　both　ends,　the　relative　L2　error　between　the　simulation　result　and　the　reference　solution　derived　from　the　Generalized　Finite　Difference　Method　(GFDM)　is　1.　15×10-2.　Furthermore,　an　in-depth　analysis　on　the　influence　of　the　hyperparameter　σ　is　carried　out,　so　the　relationship　between　σ　and　the　frequency　of　the　eigenvector　can　be　achieved.　The　analysis　results　disclose　that　through　the　selection　of　the　proper　hyperparameter,　FF-PINN　can　better　learn　vibrations　with　specific　frequencies,　so　the　dynamic　response　of　pipeline　conveying　fluid　during　vibration　can　be　effectively　captured.　This　proposed　approach　offers　an　efficient　solution　to　simulating　spatiotemporal　multiscale　structural　vibrations　with　high-　and　low-frequency　superposition.　©　2025　Chinese　Society　of　Civil　Engineering.　All　rights　reserved.