The　rapid　expansion　of　dockless　bicycle-sharing　systems　has　presented　challenges　in　efficiently　scheduling　bicycles　due　to　their　uneven　distribution　across　time　and　space.　The　accurate　prediction　of　shared　bicycle　demand　is　crucial　for　optimizing　the　scheduling　of　dockless　bicycle-sharing　systems.　This　study　introduces　the　Convolutional　and　Gated　Attention　Spatio-Temporal　Network　(CGA-STNet),　which　incorporates　multiple　spatial　features　and　time　periodicity.　The　model　effectively　identifies　the　spatial　features　of　shared　bike　orders　using　a　multi-dimensional　space　extractor.　In　the　time　processing　module,　the　Fourier　transform　is　utilized　to　extract　time　periods　and　amplitudes　from　the　data.　By　transforming　time　series　into　a　two-dimensional　matrix,　a　Convolutional　Neural　Network　is　employed　to　extract　features　within　and　between　periods.　The　experimental　results　show　that　CGA-STNet　outperforms　the　baseline　model　in　nearly　all　samples　in　both　Beijing　and　Shenzhen.　In　comparison　to　the　benchmark　model　with　the　highest　accuracy,　the　mean　square　error　of　the　CGA-STNet　model　is　reduced　by　an　average　of　16.3%　across　the　four　datasets　in　the　two　cities.　Additionally,　a　bicycle　inventory　recognition　algorithm　is　proposed　to　validate　the　practical　value　of　the　prediction　results.　The　proposed　method　enhances　the　accuracy　of　shared　bicycle　demand　prediction,　providing　valuable　insights　for　improving　the　efficiency　of　shared　bicycle　systems.　©　2024　Elsevier　Ltd