To　improve　the　reasonable　allocation　of　network　resources,　and　reduce　service　interruption　or　overwhelming　delay　for　software-defined　networking　(SDN)　due　to　excessive　network　load,　this　paper　proposes　a　network　load　balancing　scheme　based　on　gated　recurrent　units　(GRU)　aiming　at　optimizing　load　balancing　strategies.　First,　a　link　prediction　and　priority　adjustment　model　is　constructed　to　describe　the　network　load　balance　optimization　problem.　Second,　a　GRU-based　network　traffic　prediction　algorithm　is　proposed　to　capture　network　traffic　features.　Specifically,　a　graph　convolution　network　(GCN)　is　utilized　to　learn　the　spatial　features　in　the　network　topology,　and　GRU　is　used　to　capture　the　long-term　dependencies　in　the　traffic　data　sequences.　Finally,　a　logical　ordering　and　threshold　filtering　strategy　is　designed　to　achieve　priority　ordering　of　links　and　determine　the　preferred　links　for　services　to　improve　network　load　balancing.　The　experimental　results　show　that　the　GRUbased　network　traffic　prediction　algorithm　exhibits　3%　higher　prediction　accuracy　compared　to　the　existing　baseline　methods,　and　the　network　load　balancing　scheme　based　on　GRU　performs　4　x　network　load　balancing.　©　2024　IEEE.